The western honey bee, Apis mellifera, is a social insect that lives in colonies of tens of thousands, communicates the location of food through a dance, and has become the most intensively studied insect in the world.
A honey bee has roughly a million neurons — about a hundred-thousandth of a human brain — and does things with them that took researchers decades to believe. It tells its nestmates the direction and distance of a flower patch by dancing in the dark. It can be trained to treat an empty card as the smallest quantity. It sleeps, and if you keep it awake its directions get sloppier. It is also the subject of one of the most instructive arguments in the history of animal behaviour: the dance language was doubted for thirty years, and settling it required gluing radar transponders to individual insects. This page keeps the well-established findings and the genuinely unsettled ones clearly apart, because the difference between them is the interesting part.
In-depth record · 90% complete · reviewed 2026-08-09
What this page covers
One species, Apis mellifera, with many subspecies. Nearly all famous "bee" research is on this animal or on the buff-tailed bumblebee.
Often confused with: Bombus terrestris (buff-tailed bumblebee — larger, furrier, a different genus); Vespula vulgaris (common wasp — hairless, narrow-waisted, not a bee); Osmia bicornis (red mason bee — solitary, no colony, no honey)
Quick facts
Species
Apis mellifera — one of around 20,000 bee species
Colony
Typically 20,000–60,000 workers, one queen
Neurons
Roughly 1 million (a human has around 86 billion)
Communication
Waggle dance encoding direction and distance
Number sense
Orders an empty set below one
Foraging range
Commonly up to 5 km from the hive, occasionally much further
Most "bee facts" are honey bee facts, and the difference matters.
There are more than twenty thousand bee species. The great majority are solitary: no colony, no honey, no queen, often a single female provisioning a burrow. When a headline says "bees", it almost always means the western honey bee — a managed, highly social animal that is unrepresentative of bees as a whole.
Honey bee (Apis mellifera): perennial colonies of tens of thousands, stores honey, dances to communicate.
Bumblebees (Bombus): annual colonies of a few hundred, no dance, and the species used in most laboratory learning experiments.
Solitary bees such as mason and mining bees: no colony at all, and the majority of bee species and of wild pollination.
Living in a colony with overlapping generations, cooperative brood care and a division into reproductive and non-reproductive individuals. Honey bees are eusocial; most bees are not.
The most famous result in insect behaviour — and the thirty-year argument about whether it was real.
A forager returning to the hive walks a figure of eight on the vertical comb, waggling through the middle of it. The angle of that waggle run relative to straight up corresponds to the direction of the food relative to the sun. How long the waggling lasts corresponds to how far away it is. Karl von Frisch worked this out over four decades and shared a Nobel Prize for it in 1973.
Honey bees tell each other where food is by dancing
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Returning Apis mellifera foragers perform a waggle run whose angle relative to gravity encodes direction relative to the sun and whose duration encodes distance. Recruits fly the encoded vector.
Who this applies to
western honey bees
Studied in
Apis mellifera
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
Described over decades, seriously challenged for thirty years, and then confirmed directly by radar tracking of individual recruits — including a displacement condition that a scent-following explanation cannot survive.
How far it can be extended
All Apis species studied dance, though the form varies — some dance on a horizontal surface in the open rather than on a vertical comb in the dark. Bumblebees and solitary bees do not dance at all.
Caveats
Recruits use the dance to get close, then switch to scent and visual search for the final approach.
Dance dialects differ between honey bee subspecies, so distance encoding is not universal even within the species.
Where researchers disagree
From the 1960s onward, Adrian Wenner and colleagues argued that recruits located food by odour rather than by decoding the dance. The argument was substantive and unresolved for roughly three decades, until individual flight paths could be tracked directly.
Still unanswered
How precisely do recruits decode distance, given the variability between individual dances?
Radar-tracked recruits flew the dance vector, and bees released from a displaced site flew that same vector into empty ground — which rules out scent-following.
What is usually left out is that this was seriously disputed. From the 1960s, Adrian Wenner and colleagues argued the recruits were simply following the smell of the food, and that the dance was incidental. The argument was substantive, it lasted about thirty years, and it was only settled when technology arrived that could follow one bee.
How we know
Following a bee by radar
When a bee watches a waggle dance, does it actually use the information — or is it just following a scent?
For thirty years there was a genuine argument about whether the famous waggle dance conveys anything, or whether recruits simply follow the smell of the food. The test needed a way to watch where a single bee actually flew. Researchers glued a tiny transponder to bees that had just watched a dance, then tracked them with radar. Some were released at the hive as normal; others were carried to a completely different release point first.
What happened
Bees released normally flew the direction and distance encoded in the dance, then began searching for the flowers. Displaced bees flew exactly the same vector from the wrong place — and ended up over empty ground.
What it shows
The dance carries usable direction and distance information, and recruits fly it. If they were tracking an odour, the displaced bees would have found the food anyway; instead they flew a heading that only made sense from the hive.
What it does not show
It does not mean scent is irrelevant — bees clearly use it for the final approach. And this is the western honey bee; other bee species solve the problem in other ways, and dance dialects vary even between honey bee subspecies.
The controls — what makes this evidence rather than a story
Bees released at the hive gave the normal case for comparison.
Displaced bees had the same dance information but the wrong starting point.
Individual flight paths were recorded continuously, rather than only arrival at the feeder.
One well-supported result, one contested one, and the difference between them.
Numerical ability in bees has produced both a genuinely convincing finding and a widely repeated one that has not held up as well. NatureHQ separates them rather than reporting "bees can do maths" and leaving it there.
Honey bees treat "nothing" as the smallest quantity
Well supported
Good evidence backs this, though some details remain open.
Honey bees trained on a "choose fewer" rule extended it to an empty set without training, and did so more reliably as the comparison quantity increased — a distance effect consistent with placing zero at the low end of a numerical continuum.
Who this applies to
western honey bees, individually trained
Studied in
Apis mellifera
Why we rate it this way, and what the caveats are
Well supportedModerate confidence
The distance effect is the persuasive part: a simple "prefer the emptier card" rule would not produce it. Confidence is moderate rather than high because low-level visual cues are difficult to eliminate completely and replication is limited.
Caveats
Trained insects in a rewarded task, not spontaneous behaviour.
This is a sense of quantity, not a symbolic concept of zero.
Still unanswered
Does the empty-set ordering appear without extensive training?
Earlier evidence that bees use number-like information, in this case counting landmarks en route.
How we know
Showing a bee nothing at all
Can an insect treat "nothing" as a quantity — the smallest one — rather than as an absence?
Bees were trained to fly to whichever of two cards showed fewer shapes, earning sugar water for the right choice. Once they were reliable, they were shown something new: a completely blank card, against cards with one, two, three or more shapes. Nobody had ever rewarded them for choosing blank.
What happened
Bees chose the blank card more often than chance. They were also better at it when the comparison number was larger: blank-versus-five was easier for them than blank-versus-one.
What it shows
That last detail is the interesting one. If the bees were following a rule like "go for the emptier-looking card", the difficulty would not depend on the number. The fact that it does is what you would expect if they were placing the empty card at the bottom end of a number line — treating zero as a quantity.
What it does not show
It does not mean bees have a concept of zero in the mathematical sense, or that they could use it symbolically. These were trained, food-motivated insects performing a trained comparison, and their accuracy was above chance rather than near-perfect.
The controls — what makes this evidence rather than a story
Shape size, colour and arrangement were varied so that total ink or area could not be used as a shortcut.
Test trials were unrewarded, so the bees could not learn during the test itself.
Comparisons ranged across several numbers, allowing the pattern of errors to be examined.
The same research group, three years later, and a much shakier result.
Whether honey bees can add and subtract is not settled
Contested
Researchers actively disagree, and the disagreement is substantive.
One study reported above-chance performance by fourteen honey bees on colour-cued addition and subtraction of one. Published critiques argue that the training procedure and stimulus design leave simpler visual strategies available, and independent replication is limited.
Who this applies to
western honey bees, in one laboratory study
Studied in
Apis mellifera
You may have heard
“Bees can do maths.”
This rests almost entirely on a single study of fourteen trained insects whose methods have been publicly criticised. The better-supported finding from the same group is about ordering an empty set, which is a real and interesting result — and a different one.
Why we rate it this way, and what the caveats are
ContestedModerate confidence
We are confident about the state of the argument rather than about the answer. The original result is published in a reputable venue; the methodological criticisms are specific and substantive, and the field has not converged.
Caveats
Fourteen bees, roughly a hundred training trials each.
Performance was above chance, not accurate.
Where researchers disagree
Researchers have published critiques arguing that the bees could have solved the task using low-level visual cues or a simpler learned rule, and that the statistical support for arithmetic specifically is weaker than the presentation suggests.
No independent laboratory has published a clear replication.
Still unanswered
Would a design that removes correlated visual cues still produce the effect?
The zero result from the same group is better supported, and shows the numerical claims are not uniformly weak.
Honey bees use the number of landmarks they pass as one cue to distance
Well supported
Good evidence backs this, though some details remain open.
Honey bees trained to a feeder past a row of identical landmarks shifted their search location when the number of landmarks changed, indicating landmark count contributes to distance estimation alongside flight distance.
Who this applies to
western honey bees
Studied in
Apis mellifera
Why we rate it this way, and what the caveats are
Well supportedModerate confidence
A clean early result, though number and total distance are inherently difficult to separate and the effect was secondary to flight distance.
Caveats
Small numbers of landmarks only.
A serial or rhythmic cue cannot be fully distinguished from counting.
Still unanswered
Is the same mechanism involved as in the later numerical-cognition work?
The tendency for two quantities to be easier to tell apart the further apart they are. Its presence is a signature that an animal is treating quantities as points on a scale.
A honey bee routinely forages kilometres from home and returns to a hive entrance a few centimetres across. How it does that is still argued about at the level of mechanism.
Whether bees carry a map-like memory of their landscape is still argued about
Contested
Researchers actively disagree, and the disagreement is substantive.
Radar tracking of displaced honey bees shows homing from unfamiliar release points, interpreted by some researchers as evidence of a map-like spatial memory and by others as explicable by remembered panoramic views combined with path integration.
Who this applies to
western honey bees
Studied in
Apis mellifera
Why we rate it this way, and what the caveats are
ContestedModerate confidence
Both sides have published in the same top journal and neither has conceded. The observations are agreed; the disagreement is about what mechanism is required to explain them.
Caveats
A model that can reproduce a behaviour does not establish that the animal uses that mechanism.
Radar tracking constrains both terrain and sample size.
Where researchers disagree
View-based models of insect navigation account for much of the displacement data using memory for panoramas rather than a map, and their proponents argue that parsimony favours them.
Proponents of map-like memory argue the shortcuts taken from novel sites are difficult to reproduce with view matching alone.
Still unanswered
What experiment would decisively separate a map-like memory from sophisticated view matching?
Honey bees can be trained to tell photographs of human faces apart
Well supported
Good evidence backs this, though some details remain open.
Free-flying Apis mellifera learned to discriminate between photographs of human faces for a sugar reward, retaining the discrimination for up to two days, using general visual pattern mechanisms rather than face-specific processing.
Who this applies to
western honey bees, with photographs in a laboratory
Studied in
Apis mellifera
You may have heard
“Bees recognise human faces.”
They can be trained to tell particular photographs apart, and those photographs happened to be faces. A bee does not know you, and there is no evidence it treats a face as anything other than a pattern. The genuinely surprising part is that a brain of under a million neurons manages the discrimination at all.
Why we rate it this way, and what the caveats are
Well supportedHigh confidence
A straightforward, well-controlled discrimination experiment. The result itself is not in doubt; what it means is routinely overstated.
Caveats
The bees learned a pattern discrimination. Nothing suggests they perceive a face as a face.
Photographs, fixed lighting, close range and a sugar reward — none of which resemble a garden.
Still unanswered
How much of the discrimination survives changes in viewpoint or lighting?
A honey bee worker dies when it stings you — because the barb was built for insects
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
The honey bee worker sting bears backward-facing barbs that lodge in elastic tissue. Withdrawal tears the sting apparatus, venom sac and part of the abdomen from the bee, which dies within hours. Insect cuticle is inelastic, and the sting withdraws cleanly from it. Bumblebees, wasps and honey bee queens have effectively unbarbed stings and can sting repeatedly.
Who this applies to
western honey bee workers stinging mammals
Studied in
Apis mellifera
You may have heard
“Bees sacrifice themselves to protect the hive”
Sacrifice implies the outcome is the point. The barb evolved against other insects, whose cuticle releases it cleanly; mammal skin holds it, and the bee is killed by a weapon working as designed on the wrong animal. "Bees" is also wrong — bumblebees, wasps and most of the 20,000 other bee species sting repeatedly and survive.
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
Sting morphology and its consequences are directly observable and long documented; the mechanism is anatomy rather than inference.
How far it can be extended
Barbed worker stings occur across Apis. The contrast with bumblebees, wasps and solitary bees is a difference between groups, not within them.
Caveats
The bee is not sacrificing itself deliberately; autotomy of the sting is a side effect against the wrong target.
NatureHQ gives no medical guidance on stings or allergic reaction; consult a qualified clinician.
Sting pain rankings in wide circulation are subjective and self-reported.
Still unanswered
How much does losing workers to stinging actually cost a colony over a season?
Places the sting within worker anatomy and colony defence.
The sting of a honey bee worker carries backward-facing barbs. Against another insect, whose cuticle is stiff and inelastic, it goes in and pulls out cleanly, and the bee stings again. Against mammal skin, which is elastic and closes around it, the barbs hold; the bee tears free and leaves the sting, venom sac and part of its abdomen behind, and dies within hours.
This is not self-sacrifice, and describing it that way gets the biology backwards. The bee is killed by a weapon working exactly as designed against the wrong animal. A honey bee queen, whose sting is smooth, can sting repeatedly and survive — and so can bumblebees, wasps and the great majority of the world’s twenty thousand bee species.
Safety
What NatureHQ does not tell you about stings
This page explains why the sting detaches. It gives no first-aid or allergy guidance, because sting reactions range from trivial to life-threatening, treatment advice varies by country, and a nature reference is the wrong place to get it. Anyone with a known venom allergy, or anyone experiencing difficulty breathing, swelling away from the sting site or faintness, should seek emergency medical care immediately rather than reading further.
Where this applies: global
When to get help: Emergency services, or a qualified clinician, for any systemic reaction to a sting.
Words used here
Autotomy
Losing a body part as part of how something works — here, the sting apparatus tearing away.
A hive holds its brood nest near 35 °C all year — a warm-blooded object made of cold-blooded animals
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Honey bee colonies maintain brood nest temperature close to 35 °C across a wide range of ambient temperatures. Individual workers generate heat by decoupling the flight muscles and shivering, and cool the nest by fanning and by evaporating collected water. Participation is unevenly distributed across individuals, and no bee monitors the colony temperature as such.
Who this applies to
western honey bee colonies in temperate climates
Studied in
Apis mellifera
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
Directly measured by infrared thermography at both colony and individual level, and consistent with decades of hive temperature records.
How far it can be extended
Brood thermoregulation occurs across Apis, with the target temperature and the mechanisms conserved.
Caveats
Observation hives differ thermally from full colonies in tree cavities.
Brood raised a degree or two off target develops differently, so this is a requirement rather than a comfort behaviour.
Bumblebee and solitary bee nests are thermally quite different; this does not describe bees generally.
Still unanswered
What makes some individuals heat far more than others?
How does a colony trade thermoregulation against foraging in a bad season?
Colony life cycle and the developmental sensitivity of brood to temperature.
A honey bee is an insect and takes the temperature of its surroundings. A honey bee colony does not. The brood nest is held close to 35 °C through a Vermont winter and an Arizona summer alike, and brood raised even a degree or two off that develops differently — so this is a requirement rather than a comfort.
Heating is done by individual bees decoupling their flight muscles from their wings and shivering, which lets them run the largest muscles they have as a furnace while standing still. Cooling is fanning, plus water carried in and spread to evaporate. Infrared imaging shows the effort is very unevenly distributed: some bees heat hard and others barely at all.
No bee is regulating the colony temperature. Each responds to conditions where it is standing, and 35 °C is what the sum of those responses produces.
A summer worker lives about six weeks and does a different job in each stretch of it. She cleans cells for the first days, feeds larvae for about a week, then builds comb, then guards the entrance, and only in her final fortnight flies out to forage. Foraging is the dangerous job and it is given to the bees with least life left.
The schedule is not fixed. Remove the foragers and young bees switch early; remove the nurses and foragers can revert. Age sets a default that the colony’s needs override.
The three kinds of bee in a hive
Bee
Lifespan
Role
Summer worker
About 6 weeks
Every task, in age order
Winter worker
Several months
Cluster, generate heat, survive to spring
Queen
Years
Lays; mates once on a small number of flights
Drone
Weeks
Mates, or is expelled before winter
One scope warning applies to this whole page and is easiest to state here. All of it describes Apis mellifera. A bumblebee colony lasts a single season and dies, leaving only new queens; most of the world’s bees are solitary, with no colony, no honey, no queen and no workers. "Bees do X" is nearly always a claim about one domesticated species.
Words used here
Temporal polyethism
Doing different jobs at different ages. The organising principle of a honey bee colony’s workforce.
Honey is not stored nectar — bees chemically change it, then dry it out
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Foragers carry nectar in a crop separate from the digestive stomach; enzymes from the hypopharyngeal gland, including alpha-glucosidase and glucose oxidase, split sucrose into glucose and fructose and generate hydrogen peroxide. Workers then reduce water content from roughly 70% to under 20% before capping the cell.
Who this applies to
the western honey bee
Studied in
Apis mellifera
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
Enzyme production has been measured directly across worker age classes, and honey composition — sugar profile, water content, acidity, peroxide activity — is among the most thoroughly analysed of any natural product.
How far it can be extended
All Apis species store honey and share the glandular anatomy; the enzymology has been characterised in Apis mellifera.
Caveats
"Bee vomit" is a poor description: the crop is a separate storage organ, not the digestive stomach.
Honey not spoiling is explained by low available water, acidity and peroxide — not by any medicinal property.
Composition varies substantially with floral source, so single figures are averages.
Still unanswered
How much does the microbial community in the crop contribute to ripening?
Bogdanov et al., 2008 · Journal of the American College of Nutrition
Composition, water content, acidity and the basis of honey’s antibacterial activity.
A forager drinks nectar into a crop — a storage chamber ahead of the valve leading to her actual stomach, so the nectar is not digested. Back at the hive she passes it mouth-to-mouth to a house bee, and often through several bees in turn. Each transfer adds enzymes and exposes the droplet to the hive’s warm, moving air.
Two changes happen. Chemically, an enzyme splits sucrose into glucose and fructose, and a second enzyme sets up a slow reaction that releases hydrogen peroxide. Physically, the water is driven off: nectar arrives at around 70% water and honey is capped below 20%. Bees speed this by spreading droplets in thin films across cell walls and fanning air through the hive.
That combination is why honey does not spoil. There is too little available water for microbes to grow, it is acidic, and the peroxide system adds a further check. None of that makes honey a medicine, and the distinction matters: the reasons honey keeps are physical and unglamorous, and they are not evidence for the health claims that attach to it.
Nor is honey "bee vomit", a description that circulates widely. The crop is a dedicated storage organ, not the digestive stomach, and the material never enters the gut. A colony needs a substantial store to survive winter — tens of kilograms in a temperate climate — which is what the whole apparatus exists for.
Words used here
Crop
A storage chamber in a bee’s foregut, ahead of the valve into the true stomach. Nectar is carried here and is never digested.
Trophallaxis
Mouth-to-mouth transfer of liquid food between bees. It moves nectar, adds enzymes, and spreads information about what the colony has.
Bees do not build hexagons — they build rough circles that settle into hexagons
Well supported
Good evidence backs this, though some details remain open.
Wax is secreted from abdominal glands and worked at around 40°C. Cells begin closer to circular and their shared walls relax towards hexagonal junctions as warm wax reaches mechanical equilibrium between neighbouring cells.
Who this applies to
western honey bee comb
Studied in
Apis mellifera
Why we rate it this way, and what the caveats are
Well supportedModerate confidence
Direct observation of cell shape during construction supports the equilibrium account. Confidence is moderate because how much of the final geometry is physics and how much is bee behaviour is still argued in the literature.
Caveats
The balance between physical relaxation and active shaping by bees is genuinely contested.
That hexagons are the most material-efficient tiling is true and is not something any bee computes.
Wax is expensive: producing it costs several times its weight in honey.
Still unanswered
How much do bees actively shape cells rather than letting the wax settle?
Cells observed becoming hexagonal as warm wax settles at cell junctions.
Wax is secreted as small flakes from glands on the underside of a worker’s abdomen, at a particular age in her life. It is expensive: producing wax costs a colony several times its weight in honey, which is why bees reuse comb for years and why a swarm moving into an empty cavity has a great deal of work ahead of it.
The hexagon is the most over-explained shape in nature writing. Bees do not measure angles. They build cells that start closer to round, in wax kept near 40°C by the cluster, and the walls between neighbouring cells settle into hexagonal junctions because that is where the forces balance. The bees supply the heat, the packing and the wax; the geometry follows.
That the result is also the most material-efficient way to tile a plane is genuinely true, and worth separating from the explanation. It is a fact about hexagons, not a calculation any bee performs.
Words used here
Wax gland
One of eight glands on a worker’s abdomen that secrete wax flakes, active for a window in the middle of her life.
It does not hibernate, and it does not die. That is what the honey was for.
Honey bees do not hibernate and the colony does not die — it clusters and burns honey all winter
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Honey bee colonies overwinter as an active cluster of long-lived winter bees that generate heat by shivering the flight muscles and consume stored honey to fuel it. Colony size cycles from a few thousand in late winter to tens of thousands at midsummer. This contrasts with bumblebees and social wasps, whose colonies die in autumn leaving only mated queens to overwinter alone.
Who this applies to
western honey bees in temperate climates, contrasted with bumblebees
Studied in
Apis mellifera, Bombus
You may have heard
“Bees hibernate in winter”
Honey bees do neither of the two things people expect. They do not hibernate and the colony does not die off: it stays awake in a tight cluster, shivering to hold the centre warm, eating the honey it spent all summer making. That is what the honey was for.
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
Basic colony biology, directly observable and measurable as honey consumption and cluster temperature through a winter.
How far it can be extended
Overwintering as an active cluster is characteristic of temperate Apis; tropical species and races behave differently.
Caveats
Winter bees are physiologically distinct from summer workers, not simply the same bees living longer.
Starvation over winter is a common cause of colony loss, which is why stored honey matters.
Bumblebee and wasp colonies genuinely do die each autumn — the contrast causes most of the confusion here.
Still unanswered
What physiologically distinguishes a winter bee from a summer worker?
Winter bees live for months rather than the six weeks of a summer worker.
A honey bee colony spends the winter awake. The bees form a tight cluster around the queen, the ones on the inside shivering their flight muscles to generate heat while those on the outside form an insulating shell, and they rotate. Fuelling that costs honey — often fifteen to twenty kilograms of it over a temperate winter, which is precisely why the colony spent the summer making so much.
Winter bees are not simply summer bees living longer. They are physiologically different, raised in autumn with larger fat bodies, and live for months rather than six weeks. The colony shrinks from tens of thousands at midsummer to a few thousand by late winter, then builds again.
Most of the confusion here comes from other insects. A bumblebee colony genuinely does die in autumn, leaving only mated queens to overwinter alone underground; so does a wasp nest. Someone who has watched a bumblebee nest end and assumes honey bees do the same has drawn a reasonable inference from the wrong species.
Starvation is a leading cause of overwinter colony loss, and it frequently happens with honey still in the hive — the cluster cannot break to reach stores on the far side in prolonged cold.
Words used here
Winter cluster
The ball of bees that forms in cold weather, generating and conserving heat. It moves slowly through the comb as stores are consumed.
A queen and a worker are genetically identical — the difference is what the larva was fed
Well supported
Good evidence backs this, though some details remain open.
Caste in honey bees is determined by larval diet rather than by genotype. Larvae fed royal jelly throughout development become queens; those switched to worker jelly become workers. Silencing a DNA methyltransferase in worker-destined larvae produces queen-like adults without any dietary change, indicating that royal jelly acts at least partly by altering DNA methylation and therefore which genes are expressed.
Who this applies to
western honey bee caste determination
Studied in
Apis mellifera
Why we rate it this way, and what the caveats are
Well supportedModerate confidence
The dietary determination of caste is established beyond dispute. The methylation mechanism rests on an elegant knockdown experiment, and later work indicates royal jelly acts through several routes, so methylation is part of the explanation rather than the whole of it.
How far it can be extended
Diet-determined caste is characteristic of Apis.
Caveats
RNA interference knockdown is partial and can act off-target.
Queen-like morphology is not the same as a fully functional queen.
Royal jelly appears to act through more than one route; methylation is one of them.
Still unanswered
Which components of royal jelly carry the signal?
How much of the caste difference is methylation and how much is nutrition alone?
Caste determination by larval feeding, and the resulting differences in anatomy and lifespan.
A queen and a worker start as the same thing. Any fertilised egg can become either, and what decides it is diet: a larva fed royal jelly throughout development becomes a queen, one switched to worker jelly after a few days becomes a worker. The genetics are identical; what differs is which genes get read.
Royal jelly is a secretion from glands in the heads of young workers, not a stored food. The experiment that showed how it works is unusually neat — instead of feeding royal jelly and watching a queen appear, researchers silenced the enzyme that chemically marks DNA and got queen-like adults on ordinary worker food.
Who is who in a hive
Caste
Origin
Lifespan
What it does
Queen
Fertilised egg, fed royal jelly throughout
Years
Lays; mates once on a few nuptial flights
Worker
Fertilised egg, switched to worker jelly
6 weeks in summer, months in winter
Everything else, in age order
Drone
Unfertilised egg — no father
Weeks
Mates, then dies; survivors expelled before winter
The queen does not rule anything, and "queen" imports a monarchy that is not there in the same way it does for ants. She lays eggs and releases pheromones that suppress worker reproduction and signal her presence. Workers decide when to replace her, and do.
How we know
Making a queen without royal jelly
Queen and worker have the same genome and differ only in what they were fed. Does royal jelly supply something that builds a queen, or does it change which genes get read?
Feeding studies alone cannot separate those two accounts, because both predict that jelly-fed larvae become queens. The experiment worked from the other end: rather than adding jelly, it removed the machinery that keeps queen genes silent. Newly hatched larvae on a normal worker diet were treated with small interfering RNA against Dnmt3, the enzyme that adds methyl groups to DNA. If the diet works by lifting methylation, then knocking the enzyme down should produce queen-like adults on worker food — a prediction the nutritional account does not make.
What happened
A majority of the knockdown larvae emerged with queen characteristics — more ovarioles and a developed spermatheca — despite never receiving royal jelly. Controls developed as workers.
What it shows
That caste in the honey bee runs through gene expression rather than through a queen-making ingredient. The larva has both programmes available; the diet determines which one is readable.
What it does not show
It does not identify which component of royal jelly acts on methylation, nor which methylated genes matter. Queen-like is not identical to queen: the knockdown adults were scored on specific characters, and nothing here shows they would function as queens in a colony. One species, and a single enzyme out of a system with several.
The controls — what makes this evidence rather than a story
Larvae treated with a non-targeting siRNA, reared identically, provided the comparison.
All treated and control larvae were on the same worker diet, so the food was held constant while the methylation machinery varied.
Adult morphology was scored on the standard queen–worker characters, including ovariole number and the presence of spermatheca, rather than on general appearance.
Bee flight was never impossible — the calculation that "proved" it used the wrong aerodynamics
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Honey bees beat their wings through an unusually short arc of roughly 90 degrees at about 230 beats per second, generating lift through unsteady aerodynamic mechanisms including a leading-edge vortex. Asked for more force they extend stroke amplitude rather than frequency, retaining a reserve. Fixed-wing steady-state aerodynamics, which the original impossibility calculation applied, does not describe flapping flight.
Who this applies to
honey bees; the aerodynamic principles are general to insect flight
Studied in
Apis mellifera
You may have heard
“Scientists proved bumblebees cannot fly”
Nobody proved anything of the sort. The story traces to a back-of-envelope calculation treating a bee wing as a fixed aerofoil in steady airflow — which is not what a wing beating 230 times a second is doing. Once unsteady mechanisms are measured the flight is unremarkable. The anecdote survives because it flatters the idea that nature outwits science, when what it actually shows is somebody using the wrong model.
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
Kinematics measured by high-speed filming and forces measured directly on a dynamically scaled robotic wing reproducing them.
How far it can be extended
Unsteady lift mechanisms have been demonstrated across many flying insects; the specific kinematics are honey-bee measurements.
Caveats
One species; wing kinematics vary considerably across bees.
The short stroke arc is genuinely unusual and its advantage is not fully settled.
Robotic models reproduce measured kinematics rather than live control.
Still unanswered
Why do honey bees use such a short stroke arc when a longer one appears available?
Altshuler et al., 2005 · Proceedings of the National Academy of Sciences
Measured the stroke amplitude and frequency and reproduced the forces on a robotic wing.
A honey bee beats its wings about 230 times a second through a short arc of roughly 90 degrees — an odd combination that leaves it a reserve it can call on when heavily loaded. The lift comes from unsteady aerodynamics, including a vortex that forms over the leading edge of the wing and stays attached, which is not something a fixed aerofoil does.
That is worth stating because of a persistent story. The claim that science proved bumblebees cannot fly traces to a rough calculation treating a bee wing as a rigid aerofoil in steady airflow. It was the wrong model, not a paradox, and the anecdote survives mainly because it flatters the idea that nature outwits science.
Five eyes: two compound eyes for form and motion, and three simple ocelli on top of the head that track light level and horizon.
Colour vision shifted into the ultraviolet — bees see UV, blue and green, and are effectively blind to red.
Smell through the antennae, sensitive enough to distinguish floral blends and to read the colony’s chemical state.
No ears as such: bees detect near-field air movement and substrate vibration, which is how a waggle dance is read in a dark hive.
A forager typically works within two to three kilometres of the nest and will range considerably further when forage is scarce — a colony effectively surveys tens of square kilometres.
How we know
Building a wing to find out how a bee stays up
Honey bees beat their wings through an unusually short arc at a very high rate. Does that produce enough lift, and if so, by what mechanism?
A live bee is too small and too fast to instrument directly, so the work went in two stages. First, hovering bees were filmed at high speed to measure stroke amplitude and beat frequency accurately, including under conditions — heavier loads, thinner air — that force them to work harder. Then the measured kinematics were reproduced at model scale on a robotic wing in mineral oil, where the forces the wing generates can be measured directly and the fluid dynamics can be observed. Scaling the experiment up rather than shrinking the instruments is what makes the forces accessible.
What happened
Bees hover with a stroke amplitude near 90 degrees at around 230 beats per second, and increase amplitude rather than frequency when loaded. The robotic wing showed the short, fast stroke generating sufficient lift through unsteady mechanisms, principally a leading-edge vortex maintained across the stroke.
What it shows
That honey bee flight is well accounted for by unsteady aerodynamics, and that the animal has amplitude in reserve while running near its frequency limit. It also disposes of the claim that bee flight is aerodynamically inexplicable: that story comes from treating a beating wing as a fixed aerofoil in steady airflow, which is the wrong model rather than a missing mechanism.
What it does not show
A robotic wing reproduces kinematics; it does not reproduce control, and how a bee stabilises itself in turbulence is a separate question. One species. And why honey bees use such a short stroke arc when a longer one appears mechanically available is not answered here.
The controls — what makes this evidence rather than a story
Bees were filmed in low-density gas mixtures as well as normal air, so the response to increased demand revealed which parameter they adjust.
The robotic wing reproduced measured kinematics rather than idealised ones, so the model was constrained by the animal.
Dynamic similarity was matched by fluid viscosity and scale, so forces measured on the model correspond to forces on the wing.
The reliable difference is not colour, which overlaps confusingly, but body shape and covering. Bees are rounded and hairy, because hair is how pollen gets carried; wasps are smooth, with a narrow waist and a more angular outline.
Bee, wasp, hornet
Feature
Honey bee
Common wasp
Hornet
Body
Rounded, densely hairy, amber-brown
Smooth, glossy, bright yellow and black
Smooth, larger, browner yellow
Waist
Thick
Very narrow
Narrow
Diet
Nectar and pollen only
Insects for larvae, sugar for adults
Insects, sugar
Sting
Barbed; dies after stinging a mammal
Smooth; stings repeatedly
Smooth; stings repeatedly
Colony in winter
Survives as a cluster
Dies; new queens overwinter alone
Dies; new queens overwinter alone
The behaviour at a picnic is diagnostic on its own. A bee visiting flowers and ignoring people is a bee; an insect investigating a jam sandwich or a beer glass in late summer is almost certainly a wasp, whose colony no longer has larvae producing the sugary secretion the adults were living on.
Pollination is largely accidental. A bee visits a flower for nectar or pollen, and grains stick to her body — helped by the branched hairs covering bees and by the electrostatic charge a flying bee accumulates. Pollen destined for the colony is packed into baskets on the hind legs, but the grains caught loosely on the body are the ones that reach the next flower’s stigma.
What makes this work rather than merely scattering pollen is flower constancy: an individual bee tends to stick with one flower species for a foraging trip, ignoring equally rewarding alternatives. Pollen therefore arrives at a flower of the same species, where it can actually be used. The interesting wrinkle is that this may be a limitation rather than a service — switching between learned flower types appears to be costly for the bee, and the plants benefit from a constraint rather than from cooperation.
Wild insects pollinate crops better than honey bees, and honey bees cannot replace them
Well supported
Good evidence backs this, though some details remain open.
Across 41 crop systems in roughly 600 fields worldwide, wild insect visitation increased fruit set universally, while honey bee visitation increased it in 14% of systems. Managed honey bees supplemented but did not substitute for wild pollinators.
Who this applies to
41 crop systems across every inhabited continent
Studied in
Apis mellifera, Bombus
Why we rate it this way, and what the caveats are
Well supportedHigh confidence
An unusually large coordinated dataset spanning six continents, testing the same comparison in each system rather than extrapolating from one.
How far it can be extended
The comparison was made across many crops and continents specifically to test generality, which is what gives the result its weight.
Caveats
"Save the bees" usually means honey bees, which are managed livestock and not endangered. The wild pollinators doing most of the work are the ones declining.
Honey bees remain agriculturally important where wild pollinators are scarce or a crop flowers all at once.
Observational across sites; conditions were not experimentally controlled.
Still unanswered
Which wild species matter most in a given system, and can that be predicted?
Do managed honey bees compete with wild pollinators for forage?
Grüter and Ratnieks, 2011 · Communicative & Integrative Biology
Why single-species foraging makes pollination work at all, and why it may be a limitation rather than a service.
Honey bees also cannot do everything. Some plants — tomatoes, blueberries, cranberries — release pollen only when a flower is vibrated at the right frequency. Honey bees cannot buzz-pollinate; bumblebees and many solitary bees can. That is one concrete reason a landscape with only managed hives is not a landscape with working pollination.
A good collective decision, made by an animal none of which is capable of making it.
How we know
Giving a swarm two perfect houses and watching it choose
A swarm choosing between two equally good nest sites should deadlock — and a swarm that splits dies. How is the tie broken?
The deadlock was staged deliberately: swarms were offered two nest boxes of identical quality, the situation least likely to resolve itself. Every scout was individually marked, and every waggle dance recorded — along with a much briefer signal that had been noticed before and not understood, a short vibratory "stop signal" delivered by one bee butting another. The record captured not just who signalled, but who they signalled at.
What happened
Scouts committed to one site delivered stop signals specifically to scouts dancing for the *other* site, at a rate proportional to their own numbers. The symmetry broke and the swarm departed intact.
What it shows
Deadlock is resolved by cross-inhibition between the groups supporting each option — the same motif found in the neural circuits underlying vertebrate decision-making. It has been arrived at twice: once by neurons in a brain, once by bees on a branch.
What it does not show
The bees are not thinking, and the parallel with neurons is structural rather than an argument that a swarm is brain-like. The scenario is also staged: naturally available cavities are rarely exactly equal. Stop signals occur in foraging too, with a different function, so the signal is not dedicated to this.
The controls — what makes this evidence rather than a story
Two identical sites remove quality as the deciding variable, so any resolution must come from the interaction itself.
Individual marking makes it possible to see which site a signaller was committed to.
Recording the recipient of each stop signal is what distinguishes cross-inhibition from general noise.
A swarm hanging from a branch has a few days to choose a cavity, and the choice matters enormously — a bad one kills the colony over winter. Several hundred scouts go looking, a dozen or more sites get advertised, and the swarm reliably picks one of the best. The obvious explanation, that the scouts somehow compare notes, is not what happens. No bee in the process ever visits two sites.
A swarm picks the best available nest site, and no individual bee ever compares two
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Scout bees advertise discovered nest sites by dancing, with dance vigour decaying over successive returns at a rate inversely related to site quality. Scouts committed to one site deliver inhibitory stop signals to scouts dancing for competing sites. The combination produces convergence on the best-advertised site without any individual evaluating more than one.
Who this applies to
honey bee swarms choosing a nest cavity
Studied in
Apis mellifera
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
The full dance record of individually marked scouts on entire swarms, plus a direct experimental demonstration of cross-inhibition in a staged deadlock. Two independent mechanisms, each directly observed.
How far it can be extended
Observed in Apis mellifera; comparable quorum-based site selection occurs in other social insects, including Temnothorax ants, though the mechanisms differ.
Caveats
A small number of swarms, in one region, with the cavity options constrained by the experimenters.
Stop signals occur in foraging too and are not dedicated to nest choice.
The neural rule producing quality-dependent decay is inferred from behaviour.
Still unanswered
What does a scout measure to judge a cavity, and how is it converted into dance duration?
How often does the process choose wrongly, and what happens when it does?
The dance itself — the signal the whole decision process is built from.
Each scout advertises only what it found, and dances for fewer circuits on each return until it stops altogether — more slowly if the site was good. Support therefore accumulates faster for better cavities without anything doing the comparing. The arithmetic does the work.
That leaves one problem: two equally good sites produce a deadlock, and a swarm that splits dies. The answer, found by staging exactly that situation, is a brief vibratory stop signal that scouts deliver to rivals dancing for the *other* site. Cross-inhibition of this kind is the same motif found in the neural circuits that underlie decision-making in vertebrates — arrived at twice, once by neurons and once by bees on a branch.
A scout that dances for a poor site gives up sooner than one that dances for a good one. That difference in persistence, and nothing else, is what selects the better home.
A swarm settles on a nest site by building a quorum of scouts at one place. What stops two equally good sites from holding it in deadlock indefinitely?
The setup deliberately created the hardest case. Swarms on an island with no natural cavities were offered two nest boxes of matched quality, so neither site could win on merit and any resolution had to come from the interaction between the scout groups. Individual scouts were paint-marked by which site they had visited, and their behaviour on the swarm cluster was recorded — in particular who delivered stop signals, the brief vibratory pulse that causes a dancing bee to fall silent, and to whom.
What happened
Scouts advertising one site delivered stop signals preferentially to scouts advertising the other, and the rate of cross-inhibition rose as the competing group grew. Deadlock broke, and swarms committed to one box.
What it shows
That the decision includes an inhibitory term as well as an excitatory one. Recruitment alone can build support for two sites at once; cross-inhibition makes support for one actively suppress the other, which is what turns a race into a decision. The same structure appears in models of decision-making in primate brains, which is a striking convergence rather than a claim that the swarm is a brain.
What it does not show
The comparison to neural decision-making is an analogy at the level of the model, not a shared mechanism. The experiment also does not show that stop signals are used this way outside the artificial two-equal-sites situation — in the field, sites are rarely matched, and quality differences may resolve most decisions before inhibition matters much.
The controls — what makes this evidence rather than a story
Two sites of equal quality, removing site merit as the explanation for which one won.
Scouts individually marked by allegiance, so the direction of each stop signal could be attributed rather than inferred.
Swarms on an island lacking alternative cavities, so recruitment to unrecorded sites did not confound the counts.
A swarm is not an attack and not a colony in trouble. It is reproduction at the level of the colony: when a hive grows crowded, the old queen leaves with roughly half the workers, and the bees left behind raise a new queen. The cloud of bees that alarms people is a colony in transit, and it is generally at its least defensive — swarming bees have no stores or brood to protect and have filled themselves with honey before leaving.
What happens next is one of the best-studied collective decisions in any animal. The swarm settles in a temporary cluster while a few hundred scouts fly out to find cavities. A scout returning from a candidate site performs a waggle dance for it — the same dance used for flowers, now advertising real estate — and the better the site, the longer she dances. Other scouts visit the sites being advertised and dance for what they judge independently, rather than simply copying. Support accumulates for good sites and decays for poor ones, and when the scouts at one site reach a quorum the whole swarm lifts off.
Two things make this remarkable and are worth stating carefully. No bee compares the sites: no individual visits more than a fraction of them, and there is no point at which a decision is taken by anyone. And the colony reliably picks close to the best available cavity — the outcome looks like deliberation without anything deliberating, which is the same shape as the ant trail result and worth reading alongside it.
Practical
If a swarm arrives in your garden
Leave it alone and keep people and pets back, but do not panic: a settled swarm is usually docile and often moves on within a day or two of its own accord. Do not spray it, and do not attempt to move it. Most regions have beekeeping associations that will collect a swarm free of charge, and a collected swarm becomes a managed colony rather than a destroyed one.
Where this applies: Swarm collection arrangements are local. National and regional beekeeping associations maintain collector lists in most countries.
When to get help: Contact a local beekeeping association rather than a pest-control service, and seek immediate medical help for anyone with a known sting allergy.
Honey bees sleep, and losing sleep makes their directions less precise
Well supported
Good evidence backs this, though some details remain open.
Honey bees have a rest state meeting behavioural criteria for sleep. Individually sleep-deprived foragers produced waggle dances with reduced directional precision compared with disturbed controls.
Who this applies to
western honey bees
Studied in
Apis mellifera
Why we rate it this way, and what the caveats are
Well supportedHigh confidence
The rest state is well characterised by standard sleep criteria, and the deprivation experiment used disturbed-but-rested controls, which is the right comparison.
How far it can be extended
Sleep-like states have been described across many insect groups, including fruit flies and cockroaches. The specific consequence for dance precision applies only to honey bees, since only they dance.
Caveats
Physical disturbance is a blunt method and may have effects beyond preventing sleep.
Still unanswered
Does bee sleep serve memory consolidation, as it appears to in vertebrates?
Klein et al., 2010 · Proceedings of the National Academy of Sciences
Sleep-deprived foragers produced measurably less precise directional signalling.
A sleep-deprived forager still dances — it is just less precise about the direction. The colony receives worse directions because one bee did not rest.
Where the evidence is stronger than the slogans, and where it is weaker.
Neonicotinoid pesticides reduce bee reproductive success, but how much depends on where you are
Well supported
Good evidence backs this, though some details remain open.
A large multi-country field experiment found negative associations between neonicotinoid seed treatments and colony or reproductive outcomes in honey bees and wild bees, with effects differing markedly between countries.
Who this applies to
honey bees, buff-tailed bumblebees and red mason bees, at oilseed rape sites in three European countries
Studied in
Apis mellifera, Bombus terrestris, Osmia bicornis
You may have heard
“Neonicotinoids are killing the bees.”
The evidence supports reduced reproductive success rather than direct mass mortality, and the effect varied enough between countries that context clearly matters. It also flattens the more concerning finding: wild bees came off worse than the managed honey bees that get the attention.
Why we rate it this way, and what the caveats are
Well supportedModerate confidence
A rare large-scale field test with real controls. Confidence is moderate rather than high because the effects were inconsistent between countries and control sites were not free of residues.
How far it can be extended
Harm has been reported across managed and wild bees in laboratory and field settings. The size of the effect varies so much with context that a single number for "bees" would be misleading.
Caveats
Country differences were large enough that a single overall figure would mislead.
Wild bees, which are less monitored and less protected, fared worse than managed honey bees.
Part-funded by agrochemical manufacturers, disclosed by the authors.
Still unanswered
Which local factors — alternative forage, disease load, climate — drive the differences between regions?
Across 33 sites in three countries, wild bee reproduction was negatively associated with neonicotinoid residues; honey bee effects appeared in two countries of three.
The finding that gets least attention is the one that probably matters most: wild bees fared worse than the managed honey bees that dominate the coverage. Honey bees are livestock, and their numbers are a poor proxy for the health of wild pollinators.
The dance was seen long before it was read, read wrongly for two decades by the person who would eventually read it correctly, and then disputed for two more by people who had a reasonable objection. It is a good illustration of how a finding becomes secure: not by being asserted more confidently, but by someone eventually designing the measurement the critics were asking for.
1788
First observation
The dance is described before anyone suspects it means anything
Ernst Spitzner records returning foragers performing circling movements on the comb, and other bees leaving for the food afterwards. He reads it as excitement spreading, not as a message.
1923
First observation
Von Frisch reads the dance — and reads it wrong
His first published interpretation is that the round dance announces nectar and the waggle dance announces pollen. The two dances are real and the distinction between them is not what he thought.
1945
Reinterpretation
The variable turns out to be distance, not food type
Feeding stations moved progressively further from the hive show that the round dance is what a forager does for a nearby source and the waggle dance for a distant one, with the waggle run encoding direction and range.
Changes how the 1923 result reads
The observations of 1923 were sound. The interpretation failed because the experiments had varied what the bees were collecting and held distance roughly constant, so the wrong variable got the credit.
1967
Landmark experiment
The dance language described in full
Von Frisch publishes four decades of work on the waggle dance; a Nobel Prize follows in 1973.
Wenner and colleagues argue that recruits could be finding food by scent carried on the dancer and in the air, and that no experiment yet distinguishes that from bees using the dance. The objection is a fair one: both accounts predicted arrivals at the feeder.
Changes how the 1967 result reads
This is worth reading as a well-formed challenge rather than an obstruction. It named the confound precisely enough that answering it required following individual recruits in flight, which nobody could do for another thirty-five years.
1985
Landmark experiment
The colony is treated as the foraging unit
Seeley’s field programme reframes the question from what an individual bee does to how a colony allocates foragers across patches, and how it stops working a patch that is no longer worth the trip.
A swarm is shown to choose a home without comparing options
Scouts advertise sites they have inspected, and a quorum at one site — not a comparison by any individual — settles the decision. No bee visits more than one or two candidates.
Recruits are tracked by radar, and the dance is confirmed
Harmonic radar follows individual recruits from the hive, showing they fly the vector the dance encoded. The controversy over whether bees use the dance information at all is effectively closed.
Changes how the 1967 result reads
Von Frisch’s conclusion was seriously contested for decades: critics argued bees found food by odour and the dance was incidental. Radar tracking answered the objection directly by following what individual recruits actually did, rather than adding more inference to the same kind of data.
A reanalysis argues the same flight paths are explained by remembered routes and panoramic views, without any map.
Changes how the 2005 result reads
This is a live disagreement rather than a settled correction, and NatureHQ carries both. The question is not whether bees navigate well — they plainly do — but whether the evidence requires a map, and the critics argue a simpler account fits the same data.
Queen and worker are traced to which genes are read
Silencing a DNA methyltransferase in larvae produces queen-like adults on a worker diet, indicating that royal jelly acts by changing gene expression rather than by supplying a substance that builds a queen.
Scouts advertising one site deliver stop signals to scouts advertising another, an inhibitory cross-talk that prevents deadlock between two good sites.
Changes how the 1999 result reads
The 1999 account explained how a swarm reaches agreement. It left open why two strong candidates do not stall it indefinitely, and the stop signal is the missing half of that mechanism.
Why it matters: It is one of the most widely repeated claims about insect intelligence, and it rests on a single study of fourteen bees whose methods have been publicly criticised.
What would settle it: An independent replication with stimuli designed to remove correlated low-level visual cues.
Do bees hold anything like a map of their landscape?
Why it matters: It is the difference between an insect following remembered views and one representing space — a genuinely different kind of mind.
What would settle it: An experiment whose outcome view-matching models cannot reproduce.
Why do neonicotinoid effects differ so much between countries?
Why it matters: Regulation is written as though the answer were the same everywhere. The best field evidence says it is not.