Sleep is not the same as being still. It is identified by four things together — a typical posture, reduced responsiveness, waking quickly, and sleeping more after being kept awake. By that standard flies and octopuses sleep. By the stricter brain-activity standard, only mammals, birds and some reptiles have been shown to.
The interesting thing about animal sleep is how much of the difficulty is definitional. Everyone knows what sleep looks like and almost nobody can say what it is, and the two available answers do not pick out the same animals. The behavioural definition asks four questions: does the animal adopt a characteristic posture or place, does it stop responding to things it would normally react to, can it be roused quickly, and — the one that does the real work — does it sleep more after being prevented from sleeping? That last criterion is what separates sleep from merely stopping, because an animal repaying a debt is regulating something. On that standard a fruit fly sleeps, a honey bee sleeps and gets worse at giving directions when it cannot, and an octopus cycles between two states one of which involves dramatic colour changes. The stricter definition asks what the brain is doing, and that has only been recorded in a much smaller set of animals. Two further things get lost in popular retellings: a dolphin does not switch half its brain off — it runs one hemisphere in slow-wave sleep while the other keeps it swimming and breathing — and sleep duration across species correlates with almost nothing anyone expects it to.
Developed record · 91% complete · reviewed 2026-08-10
What this page covers
A behavioural state rather than a group of organisms. Covers mammals, birds, reptiles, fish and invertebrates where sleep has actually been tested for.
Often confused with: Hibernation and torpor, which are metabolic states with slow arousal; Simple inactivity, which meets none of the criteria on its own
Quick facts
Four criteria
Posture, reduced responsiveness, quick arousal, rebound after deprivation
The four criteria, and why the fourth one carries the argument.
An animal that has stopped moving is not thereby asleep
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Behavioural sleep is identified by four criteria applied together: a species-typical posture or site, reduced responsiveness to stimuli, rapid reversibility distinguishing it from torpor or coma, and homeostatic regulation shown as increased sleep following deprivation. Physiological sleep, defined by electroencephalographic state transitions including non-REM and REM, is established in mammals and birds and documented in some reptiles; it is not recorded in most invertebrates, where only the behavioural criteria are available.
Who this applies to
animals tested against the behavioural criteria; physiological criteria in mammals, birds and some reptiles
Studied in
Animalia
You may have heard
“All animals sleep”
It depends entirely on which definition is being used, and the popular version quietly uses the mammalian one everywhere. Behavioural sleep is very widespread; sleep with the brain-state architecture people picture is confirmed in a much smaller set of animals. An animal that is simply still has demonstrated nothing at all.
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
The criteria are the standard framework in the field and are testable directly. The homeostatic rebound in particular is a positive experimental result rather than an observation of stillness.
How far it can be extended
The behavioural criteria have been applied across vertebrates, insects and molluscs; the physiological criteria require electrophysiology and have been applied far more narrowly.
Caveats
Absence of recorded sleep in a group usually means nobody has recorded it.
Deprivation experiments are stressful, and stress has effects that are hard to separate from sleep loss.
Meeting the behavioural criteria does not establish that the state resembles human sleep in any experiential sense.
Still unanswered
What is sleep actually for, given how differently it is distributed across animals?
Do the reptile sleep states share an origin with the mammalian ones, or arrive at them separately?
Records two-state sleep in a reptile, extending the physiological definition beyond mammals and birds — on a cycle roughly seventy times faster.
An animal that is not moving might be sleeping, resting, hiding, digesting, too cold to move, or dead. Behavioural sleep is what remains once those are excluded, and it is established by four things holding at once rather than by any one of them.
A species-typical posture or place — hanging, curled, tucked into a crevice, wedged on the seabed.
Reduced responsiveness: it takes a stronger stimulus than usual to get a reaction.
Rapid reversibility, which is what separates sleep from torpor, hibernation or a coma.
Homeostatic rebound: prevent it and the animal sleeps more afterwards, as though repaying something.
The fourth is the one that turns an observation into a finding. Anything can be still. An animal that sleeps longer after being kept awake has demonstrated that the state is regulated — that something is being tracked and made up — and that is why the fruit fly result changed the field rather than merely adding to it.
The stricter definition is physiological: distinct brain states, visible as slow waves alternating with faster low-amplitude activity, which in mammals and birds are called non-REM and REM. That requires electrodes, which is why the physiological map of sleep is small and the behavioural one is large. Absence of recorded sleep in a group usually means nobody has recorded it.
Words used here
Homeostatic rebound
Sleeping more than usual after being kept awake. The strongest single piece of evidence that a state is sleep rather than inactivity.
Torpor
A drop in body temperature and metabolism lasting hours to days. An animal in torpor cannot wake quickly, so it is not asleep.
Non-REM and REM
The two broad sleep states identifiable from brain activity in mammals and birds. Named for rapid eye movement, which accompanies one of them.
Dolphins and some birds sleep one hemisphere at a time, with the matching eye open
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Unihemispheric slow-wave sleep is recorded in cetaceans, eared seals and several bird groups: slow waves appear in one cerebral hemisphere while the other shows waking activity, and the eye opposite the sleeping hemisphere closes while the other remains open. In cetaceans the hemispheres alternate over periods of roughly one to two hours, and bilateral slow-wave sleep is essentially absent.
Who this applies to
cetaceans, eared seals and several bird groups where it has been recorded
Studied in
Cetacea, Otariidae, Aves
You may have heard
“Dolphins sleep with half their brain switched off”
Nothing switches off. One hemisphere enters slow-wave sleep — an active, structured state — while the other stays awake enough to keep the animal swimming and surfacing to breathe. The arrangement exists because a dolphin that lost consciousness entirely would stop breathing, not because half a brain is enough to get by on.
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
Directly measured by simultaneous recording from both hemispheres, with the behavioural correlate of the contralateral eye visible from outside.
How far it can be extended
Recorded independently in cetaceans, eared seals and birds; true seals and terrestrial mammals sleep bihemispherically.
Caveats
"One half switches off" is wrong: the sleeping hemisphere shows slow-wave activity, which is a state rather than an absence of one.
Whether cetaceans have REM sleep at all is unresolved, and the evidence for it is weak.
Recording requires restraint or implantation, both of which affect the animal being recorded.
Still unanswered
Do cetaceans experience REM sleep, and if not, what does that imply about what REM is for?
What determines how much of a migratory bird’s flight is spent in unihemispheric sleep?
Records both unihemispheric and bilateral sleep in a bird, in flight, over open ocean.
A dolphin that lost consciousness completely would stop swimming and stop surfacing, and breathing in a dolphin is a voluntary act rather than an automatic one. The solution is unihemispheric slow-wave sleep: one cerebral hemisphere shows the slow waves of deep sleep while the other stays awake, and the eye opposite the sleeping side closes while the other remains open. After an hour or two they swap.
The phrase "half the brain switches off" is worth resisting, because a sleeping hemisphere is not an inactive one. Slow-wave sleep is a structured, high-amplitude state that costs energy and does something; what it does is the open question. The arrangement is not half a brain making do — it is two halves taking turns at a state neither can skip.
How we know
Recording half a brain asleep while the other half swims
A dolphin has to surface to breathe, and breathing in a cetacean is voluntary. How does an animal like that sleep at all?
Electrodes placed over both cerebral hemispheres record activity simultaneously while the animal swims, with eye state filmed alongside. The simultaneity is the whole design: a single-channel recording averaged across the head would show an intermediate state and miss the asymmetry entirely, which is roughly what happened before anyone thought to record both sides at once.
What happened
Slow-wave activity appears in one hemisphere while the other shows waking patterns, and the eye opposite the sleeping hemisphere closes while the other stays open. Hemispheres alternate over roughly one to two hours. Bilateral slow-wave sleep is essentially absent in cetaceans, and REM sleep is not clearly identifiable.
What it shows
That deep sleep and continuous voluntary behaviour can be combined by dividing them between hemispheres. It also explains the swimming behaviour of sleeping cetaceans, which circle slowly with the open eye toward their group.
What it does not show
It does not show that half the brain switches off — slow-wave sleep is an active, high-amplitude state, and the sleeping hemisphere is doing something rather than nothing. It does not establish that cetaceans lack REM sleep, only that it has not been convincingly recorded, which in an animal this difficult to instrument is a weaker statement. And every recording requires restraint or implantation, both of which change the animal being measured.
The controls — what makes this evidence rather than a story
Both hemispheres recorded at the same time, so asymmetry is measured rather than inferred from an average.
Eye state filmed continuously, giving an externally visible correlate that can be checked against the electrical record.
Long recordings spanning multiple cycles, so that alternation between hemispheres is observed rather than a single snapshot.
Comparison with terrestrial mammals recorded the same way, establishing that the asymmetry is not an artefact of the method.
It is not a marine mammal speciality either. Several bird groups do it, and ducks at the edge of a sleeping row keep the outward-facing eye open more than birds in the middle — the sleeping side being the one facing safety. Frigatebirds flying continuously over the ocean for days sleep both unihemispherically and with both hemispheres at once, while circling upward in rising air.
True seals sleep with both hemispheres like a terrestrial mammal; eared seals switch to unihemispheric sleep in water and back to bilateral sleep on land. The same animal uses whichever the situation requires.
Words used here
Unihemispheric slow-wave sleep
Deep sleep in one half of the brain while the other stays awake. Found in cetaceans, eared seals and several birds.
Keeping an insect awake makes it worse at its job the next day
Well supported
Good evidence backs this, though some details remain open.
Honey bees deprived of sleep show degraded precision in the waggle dance, specifically in the directional component that recruits nestmates to a food source, without a general reduction in activity. Rest-deprived Drosophila show a homeostatic rebound and impaired performance on learning tasks. Both are evidence that the state is regulated and functional rather than merely permitted.
Who this applies to
two laboratory insect species under experimental deprivation
Studied in
Apis mellifera, Drosophila melanogaster
Why we rate it this way, and what the caveats are
Well supportedModerate confidence
The deprivation results are clear and replicated in two species. Confidence is moderate because deprivation methods are stressful and stress is a plausible partial explanation that the designs reduce rather than eliminate.
How far it can be extended
Homeostatic regulation of rest has now been demonstrated in several arthropod species, though the functional consequences have been measured in few.
Caveats
Deprivation requires disturbing the animal, so stress cannot be entirely excluded.
Two intensively studied laboratory species stand in for a very large group.
Degraded performance shows that the state matters; it does not show what the state does.
Still unanswered
Why does sleep loss affect directional precision specifically rather than activity generally?
Homeostatic rebound and pharmacological responses matching mammalian sleep.
Insects were assumed to rest rather than sleep until somebody applied the criteria properly. Fruit flies raise their arousal threshold, adopt a posture, and — the decisive result — rest substantially more after being deprived. They respond to caffeine and antihistamines the way mammals do.
How we know
Keeping a fly awake to find out whether it was asleep
Flies stop moving for hours at a time. Is that sleep, or just an insect standing still?
The design refuses to assume and tests each criterion separately. Arousal threshold was measured by delivering mechanical stimuli of graded strength to active and inactive flies and recording which produced a response. Reversibility was checked by confirming the state ended immediately on stimulation. Then the decisive manipulation: flies were kept from resting by automated disturbance overnight, and their rest was measured over the following day against undisturbed controls. Finally, caffeine and antihistamines were administered to see whether the state responded pharmacologically the way mammalian sleep does.
What happened
Inactive flies had a substantially raised arousal threshold, adopted a characteristic posture, and woke immediately when stimulated. After deprivation they rested markedly more than controls. Caffeine reduced the state and antihistamines increased it.
What it shows
That fly rest satisfies the behavioural definition of sleep, and — through the rebound — that it is homeostatically regulated rather than merely permitted. Regulation is the difference between sleep and inactivity, and it is what made the fly a usable genetic model for sleep research.
What it does not show
Deprivation requires physically disturbing the animal all night, and stress produces behavioural changes of its own that this design reduces but cannot remove. Meeting the behavioural criteria says nothing about brain states, which are not recorded here in the way they are in a mammal. And a pharmacological response shared with mammals is suggestive of a shared mechanism without demonstrating one — the same drug can act on different systems.
The controls — what makes this evidence rather than a story
Undisturbed flies handled identically apart from the disturbance itself, isolating deprivation from handling.
Arousal thresholds measured in both active and inactive flies, so the raised threshold is a comparison rather than an assertion.
Graded stimuli rather than a single intensity, so the threshold is quantified.
Locomotor activity tracked continuously, which separates a genuine rebound from a general slowing after disturbance.
Honey bees go further, because there is something measurable to degrade. A bee prevented from sleeping still forages and still dances, but the directional component of its waggle dance becomes imprecise — the part that tells nestmates where to go. The colony is being given worse information by a bee that appears to be working normally.
How we know
A tired bee gives worse directions
Sleep loss degrades performance in mammals. Is there anything measurable to degrade in an insect?
The clever part is the deprivation method. Handling bees to keep them awake would introduce stress and disturb the colony, so instead steel markers were glued to the bees and a magnet moved outside the hive at night, jostling only the marked individuals without opening the hive or touching anything. The following day the returning foragers’ waggle dances were filmed and the directional component measured against the known bearing of the feeder.
What happened
Sleep-deprived bees continued to forage and to dance, but the directional precision of the waggle run deteriorated significantly. The signal was still produced; it was less accurate.
What it shows
That sleep loss in an insect has a specific, socially consequential cost. The colony is being given degraded information by a bee that looks like it is working normally, which is a much sharper result than "tired animals do less".
What it does not show
It does not explain why direction specifically is affected while the ability to dance at all is not. It does not measure whether the imprecision costs the colony anything at the level of food collected. And magnetic jostling is still disturbance — gentler and better targeted than handling, but not the absence of stress.
The controls — what makes this evidence rather than a story
Bees carrying identical markers made of a non-magnetic metal, disturbed by nothing while experiencing the same attachment procedure.
Disturbance applied at night without opening the hive, leaving colony temperature, humidity and social conditions undisturbed.
Dance direction scored against the true bearing of a feeder at a known location, so error is measured rather than judged.
Foraging activity recorded alongside, establishing that deprived bees still flew and still danced.
Octopuses cycle between a quiet state and an active one in which the skin flickers through colour changes and the eyes and suckers twitch — superficially like REM sleep, and reached by an animal whose last common ancestor with us had nothing that could be called a brain. Whether the resemblance means anything about the state itself, or is convergence on a similar solution, is exactly the sort of question NatureHQ leaves open rather than resolving with an analogy.
Sleep duration across species varies from around two hours to around twenty, and the pattern resists every tidy explanation. It does not track brain size, intelligence, lifespan or body mass in any simple way. What it does track, loosely, is ecology: animals that can afford to be unconscious sleep more.
Reported daily sleep, and the constraint that seems to matter
Animal
Approximate daily sleep
The constraint
Brown bat
Around 19 hours
Small, safe roost, energetically expensive flight
Domestic cat
Around 12–16 hours
Predator with no predators and a high-energy diet
Human
Around 7–8 hours
Safe sleeping site, historically in groups
Elephant
Around 2–4 hours in the wild
Enormous intake requirement; must keep eating
Giraffe
A few hours, often standing
Large, exposed, slow to get up
Frigatebird in flight
Around 42 minutes
Cannot stop flying over open ocean
Two cautions about that table. Wild figures are much lower than captive ones for most species, and a great many quoted numbers come from captivity where the animal has nothing else to do. And "hours of sleep" flattens a real difference in structure: an elephant sleeping two hours in the wild is not a human sleeping badly.
The honest summary of what sleep is for is that nobody knows. The leading candidates — clearing metabolic products, consolidating memory, saving energy, enforced immobility at times when being active is dangerous — are not mutually exclusive and none of them explains why the state has to involve losing awareness of the world.
Most species-level sleep questions have a general answer, and a few have a specific one worth going to.
Dolphins and orcas: unihemispheric, continuous swimming, no confirmed REM sleep.
Sharks: many must keep water moving over the gills, so rest is a reduced-activity state rather than stopping; some species can pump water while stationary.
Bats: among the longest sleepers recorded, and hibernation in temperate species is a separate metabolic state entirely.
Honey bees: sleep is real, structured by age and role, and losing it degrades the waggle dance.
Octopuses: two alternating states, one with dramatic skin activity.
Elephants: remarkably little in the wild — a few hours, often standing, and not every night lying down.
The arc here is a definition widening under pressure. Sleep was described in humans, then found in other mammals, then in birds — and each extension beyond that required someone to ask what the word was actually claiming, and to test it rather than assume it.
1953
First observation
REM sleep identified
Aserinsky and Kleitman record periods of rapid eye movement and fast, low-amplitude brain activity during human sleep, establishing that sleep has more than one state.
1964
Modern discovery
Unihemispheric sleep found in dolphins
Recordings from dolphin brains show slow waves in one hemisphere while the other remains awake, explaining how an animal that must surface to breathe can sleep at all.
2000
Landmark experiment
A fly is shown to sleep
Rest in Drosophila is tested against the behavioural criteria — arousal threshold, reversibility, and a rebound after deprivation — and passes, opening sleep to genetic dissection.
Changes how the 1953 result reads
The mammalian definition had made sleep a question about brain states. Applying the behavioural criteria instead let the question be asked of animals whose brains cannot be recorded that way, and the answer was yes.
A review sets the behavioural criteria against the physiological ones and shows they do not select the same animals, and that sleep duration correlates poorly with the things people expect.
Bearded dragons alternate between slow-wave and low-amplitude states through the night, on an 80-second cycle rather than the 90-minute mammalian one.
Changes how the 1953 result reads
If two-state sleep exists in reptiles, the architecture predates mammals and birds — or has arisen more than once. The very different cycle length is a caution against assuming the states mean the same thing.
Why it matters: Every candidate function — waste clearance, memory consolidation, energy saving, enforced safety — explains part of the pattern. None explains why the state requires losing awareness, which is its most dangerous feature.
Do cetaceans have REM sleep?
Why it matters: If an animal with a large brain and complex behaviour manages without it, REM cannot be as essential as the mammalian literature assumes.
Is reptile two-state sleep the same architecture as the mammalian one?
Why it matters: Homology would put the origin of sleep states far deeper in the vertebrate tree; convergence would say the structure is a good solution reached more than once.
How much of an animal’s sleep debt can be repaid at all?
Why it matters: Rebound is partial in every species measured, which suggests some of what is lost is simply lost.