He hasn't lain down. Or you haven't seen him.

I could not have told you when my horse last lay down. Not roughly. At all.

Giles was in a stable at the time, on shavings I banked up round the edges and fluffed in the middle, and most mornings there was a shape pressed into the middle of it and shavings caught in his mane. I read that shape as evidence that things were fine. What I never asked was how long he had been in it, or whether it happened again after I turned the light off, or what it would mean if the shape were not there.

It turns out that last question is one of the most-searched things in British horse ownership, and one of the least well answered. Type "my horse never lies down" into Google and page one is Quora and two forum threads. Nobody authoritative has written it down.

So here is what four papers say, including one published this January, and where the research runs out. It runs out earlier than the internet suggests.

What a horse's night actually looks like

The only full synthesis anyone has is a 2022 review by Linda Greening at Hartpury and Sebastian McBride at Aberystwyth, in Frontiers in Veterinary Science. Everyone writing about equine sleep ends up citing it, because nothing else gathers the scattered literature into one place.

Total sleep is somewhere around three and a half to four hours in 24, and the figure moves depending on how you measure it. Electrode studies put it between 172 and 262 minutes. Studies that scored sleep by watching the horse instead range from 65 to 382 minutes, which tells you more about the difficulty of the question than about horses.

What matters more than the total is the shape. Human sleep arrives in four or five long blocks, with REM sequences averaging 22 minutes. A horse gets around 10 REM sequences a night averaging under four minutes each, and the longest average reported in any equine study is 5.22 minutes. Between them the horse wakes up constantly: roughly seven wake sequences per sleep cycle, most under a minute.

So a horse's night is not a shorter version of ours. It is a different substance.

Figure 1

One night, four stages

A stabled horse, 8pm to 8am. Hover or tap a band to see what the horse's body is doing. The ember bands are REM. There are 11 of them and none lasts five minutes.

8pm10pm12am2am4am6am8am

Swipe the night sideways to reach the morning.

Wake Drowsy NREM, slow-wave REM, the restorative one

Hover or tap a band to see what is happening in the horse's body. The one-minute wake bands are too narrow to tap, so open the table below to read them.

40 minTotal REM, in 11 bouts, longest five minutes

3h 40mTotal sleep across a 12-hour night

29Interruptions once the horse was asleep

Show the whole night as a table

Why REM needs him on the ground

REM is defined by muscle atonia. Skeletal tone drops to almost nothing, which is why humans do not act out their dreams, and why a horse cannot stay upright through it.

Drowsing and slow-wave sleep come easily on his feet. About three-quarters of a standing horse's sleep is slow-wave, which is real, restorative sleep, so a dozing horse is not getting nothing. What he is not getting is REM, and REM needs him down, either flat on his side or on his chest with the muzzle actually resting on the floor. Head propped up on the chest is scored as slow-wave sleep. Muzzle on the ground is scored as REM. That is the whole distinction, and it is finer than most of us have been taught.

There is one exception, and it is worth knowing because it explains a frightening thing owners sometimes see. A horse very short of REM can slide into it standing up. Neck tone goes, the head drops, the forelegs buckle, and the muzzle hits the floor, at which point the horse wakes up. Brief, rare, and self-terminating. It is also the single most visible sign that something has gone wrong.

Figure 2

What each posture allows

The line is not standing versus lying. It is whether the neck is genuinely unsupported. Keep scrolling.

• Drowsiness
• NREM, slow-wave
• REM

Standing

Stay apparatus engaged, three legs loaded, one hind rested. Around three-quarters of a standing horse's sleep is slow-wave, so a dozing horse is not getting nothing. What he cannot get here is REM, because tone drops to almost nothing in REM and he would go over.

Sternal

Lying on the chest — and this is where the distinction gets finer than most of us were taught. Down is not the test. The muzzle is. A horse can be lying on his chest with his head still held and be scored as slow-wave sleep, no differently from standing.

Lateral

Flat on the side, legs extended, neck along the ground. Full atonia. This is the posture that worries people who find their horse like it, and on the review's own figures it is the one doing the most work.

Is it normal for a horse never to lie down?

The best evidence on that question is a survey, with everything that implies. Heli Suomala and colleagues at Helsinki asked 1,749 owners in Finland, Sweden and Norway one deceptively simple thing: do you think your horse has problems related to rest or sleep? Of those, 88 said yes. Then they looked at what the 88 had in common.

Two numbers dominate. Owners whose horse had lost its balance or gone down unexpectedly in the past year were 40 times more likely to suspect a sleep problem. Owners who had never seen their horse lie down were 28 times more likely.

That second sentence needs reading twice, because it is not what it first appears to be. It does not say that horses who never lie down are 28 times more likely to have a sleep problem. It says that owners who have never watched their horse lie down are 28 times more likely to think something is wrong. Both halves come from the same person filling in the same form on the same afternoon, so a portion of the association is built in by construction. The authors knew that, which is why they put this variable in a model they labelled consequences rather than causes.

And the confidence interval runs from nine to 89, which is a polite way of saying the estimate rests on 32 animals: 16 whose owners were worried, 16 whose owners were not.

None of which makes it useless. A lower bound of nine is still a large effect, and the finding has a plain mechanical reading that does not depend on the statistics at all. If nobody has ever seen the horse lie down, nobody knows whether he does.

Figure 3

What raises the odds, and by how much

"Three times more likely" means suspected sleep disturbance was three times as common among horses with that factor. It does not mean your horse has a one-in-three chance of anything.

What may cause itThe authors' first model: management and horse factors.

8.6×more likely Hard and wet lying surfaceThe knee test: it hurt to kneel, and your knees came up wet after 15 seconds. Interval 2.4 to 31, and only 14 horses in the whole study fell in this class.

3.5×more likely Hard but dry lying surfaceInterval 1.6 to 7.8. Hardness is common to both significant terms; wetness compounds it.

3.1×more likely Currently ill or injuredInterval 1.9 to 5.2. Getting down and up again is physically expensive.

1.12×per year of age Older horsesSmall each year, and it compounds. A 20-year-old carries meaningfully higher odds than a 10-year-old.

1.05×per cm of height Taller horsesInterval 1.03 to 1.08. The bigger the horse, the more work it is to stand up.

What you can see once it existsA separate model. These numbers are not interchangeable with the ones above.

40×more likely Lost balance or fell unexpectedly in the past yearThe largest number in the paper. Interval 17 to 97, based on 92 horses.

28×more likely Never seen lying downInterval 9 to 89, resting on 32 horses. Note the variable is "not observed lying down", which is a fact about the watching as much as about the horse.

16×more likely Struggles to lie down or get upInterval 6 to 44.

9.2×more likely Unexplained overnight injuriesInterval 3.4 to 25. Fetlock scrapes with no obvious cause.

3.9×more likely Sleepy during the dayInterval 1.7 to 8.8.

Show both models as a plot, with the full intervals

How long can a horse go without lying down?

Nobody knows.

Not approximately, not within a range, and not in any published source I could find. Greening and McBride say it themselves, at the end of a review of the entire literature: it is "not known at what point sleep deprivation becomes chronic". Joe Bertone, who has seen more of these cases than almost anyone, puts it in his own proceedings as "several days" and leaves it there.

Which is odd, because a number circulates freely. Search for it and you will be told that a horse can manage three or four nights without lying down, and that trouble starts somewhere between seven and 14 nights. I have seen that figure repeated as though it were settled.

Here is where it comes from. The most prominent source for it is an article on the website of an American supplement company, which quotes a veterinary researcher saying that problems begin when a horse goes "seven to 14 nights without proper REM sleep". The same article states that 90% of sleep-deprived horses will be injured, attributed to "a study", with no citation given. Neither figure appears in that researcher's own published work.

The number 14 does appear in the equine sleep literature, a few lines away from equine material in the Greening and McBride review. It is human. It comes from a 2003 study of people put on 14 consecutive nights of restricted sleep, and it is a definition of chronic deprivation in humans rather than a threshold for horses. That is almost certainly the whole provenance.

One more claim in this family deserves the same treatment. The January 2026 Helsinki paper states that horses can suppress REM sleep entirely for up to several weeks, and cites a 1986 review in Veterinary Clinics of North America. A review, not a study. Whether the figure originates there is not something I could establish, because the full text is behind a paywall and no open copy exists.

So the honest position is this. Recumbent sleep deprivation is real, it is documented, and it ends in collapse. How many nights it takes is not known, and everyone who tells you otherwise is quoting a supplement company or a study about people.

What stops a horse lying down

The Suomala data points at four things: pain, size, age, and the surface.

The surface is the one an owner can change tonight, and it is worth being careful about what the study actually found. Softness and dryness were not tested separately. They were combined into one measure, taken by kneeling on the bed where the horse lies and asking two questions: does it hurt, and are your knees wet after 15 seconds. Hard and dry came out at 3.5 times the odds. Hard and wet came out at 8.6.

Depth never made it into the model at all. It was tested crudely, as more or less than 5cm, found nothing, and was dropped. Helsinki's own press release nonetheless says the amount of bedding matters more than the type, which is not what their paper reports.

But the wider literature does support depth, and this is where the survey and the experiments part company. Greening and McBride, reviewing the studies that actually manipulated bedding rather than asking owners about it, report that depths under 10cm significantly reduce how often a horse lies down at all, whatever the material. The experiments changed something. The survey only asked.

So the banked-up bed was not sentiment. It was closer to right than the survey suggests, and the reason to keep doing it is that under 10cm the horse stops going down.

Then there is everything that breaks a night up without shortening it. In 2023 Greening and colleagues published the first serious attempt at measuring sleep quality in horses, in Biology Open. With no self-report available they borrowed the other trick from human sleep science and counted the interruptions: state duration divided by the number of short awakenings. Five indices rather than one number, and the principle under all of them is that a night in pieces is not the same night as one that ran through, even when the totals match.

Figure 4

Same total REM, different night

Both horses got 40 minutes. Look at the shape of it before you reveal the arithmetic.

Horse A, 11 bouts

REM quality index:3.3

0h4h8h

Horse B, 24 bouts

REM quality index:1.6

0h4h8h

Reveal the arithmetic

Which means the observation I had been quietly pleased about for years, that he lay down four times last night, might have been the wrong thing to be pleased about.

What people call lazy

In January 2026 Mira Hämäläinen, Iina Brotherus and colleagues at Helsinki published the first field test of whether the way a horse sleeps predicts how he behaves while learning something.

The kit list is the best thing about it: two researchers, two upturned rubber buckets, three car tyres to stop the buckets wandering, and a kilo of carrot slices per horse per phase. There were 16 riding-school horses, mean age 15, tested in their own paddocks rather than a laboratory. Nights filmed across five sessions of two consecutive nights each over six weeks, with REM-like sleep scored from the footage by a single rater who did not know which horse she was watching or what bedding it was on.

Figure 5

The spatial reversal learning test

Two buckets, one rule, and a rule that keeps changing. The apple marks the correct choice at each stage.

🍎

Horse learns: left means treat

🍎

Reversal 1

Hits threshold, seven correct in a minute

🍎

Rule flips, now right means treat

🍎

Reversal 2

Hits threshold again

🍎

Rule flips back, left again

Each time a horse learns the new rule and hits the threshold, that is one completed reversal. The test keeps flipping. The question is not whether the horse gets it right. It is how many rounds he will keep trying before he walks off.

What they found

123456 78910111213141516

Walked away (6) Kept going (10)

24 minMean REM-like sleep, the six who quit

59 minMean REM-like sleep, the 10 who kept going

REM duration did not predict errors. It predicted willingness to keep trying.

The horses who quit were not making more mistakes. Error rate showed no relationship with sleep at all. What sleep predicted was how long a horse would keep engaging with a task that kept moving the goalposts.

A test that counts only mistakes misses the thing that matters most in a horse's working life, which is whether he is still willing.

There is a second study worth putting next to it, because it comes at the same question from the opposite direction and nobody talks about it. In 2024 a Brazilian group ran 10 purebred Lusitano dressage horses through a crossover design in which recumbency was prevented for 72 hours. Performance on a spatial memory task tended to get worse, not better, at p = 0.084. That is 10 horses and a non-significant result, so hold it lightly. But it is the only experiment anyone has run on REM restriction in competition horses, and it points the same way as Helsinki.

The collapse that usually isn't narcolepsy

There is a thread running through the veterinary literature that gets less attention than it should, about a diagnosis handed to adult horses who fall over.

True narcolepsy is neurological, linked in humans and dogs to a shortage of the neuropeptide hypocretin. There is a recognised hereditary form in miniature horses and in certain Lipizzaner lines, and it shows up in foals. UC Davis puts it plainly: narcolepsy is uncommon in horses, and episodes that look like it are more likely to come from sleep deprivation, whether because pain stops the horse lying down or because light and noise stop him sleeping.

The best evidence for that is a doctoral thesis, which is why almost nobody cites it. In 2017 Christine Fuchs at Ludwig-Maximilian University in Munich advertised in a horse magazine for owners of horses diagnosed with narcolepsy. She got 177 replies and examined 39 horses on site, with 24-hour video, polysomnography, clinical examination and bloodwork. The collapse episodes averaged around 60 a day. In the worst cases they reached 199. Horses who lay down and rolled had significantly fewer of them, and her conclusion was that most of these horses were not neurologically ill. They were short of lying-down sleep, and the cause was in the housing.

Joe Bertone, emeritus professor at Western University of Health Sciences, has spent 20 years collecting these cases. He presented on 876 of them at the National Alliance of Equine Practitioners conference in Saratoga Springs in September 2024, and has said he has seen closer to 1,100 since. His position, as reported, is that no adult horse he has examined has met the criteria for true narcolepsy.

That figure deserves a caveat, and it is not a small one. The 876 cases have never been published. They exist in trade coverage of a talk, with no case definition, no methods and no denominator available to check. Bertone's peer-reviewed record on this runs to a textbook chapter and a 2006 case report describing two horses. The jump from two published cases to 1,100 spoken from a podium is the kind of gap that would normally make me leave a claim out. I have kept it because it is the only clinical caseload anyone has, and because a clinician saying "I have never once seen the real thing" is information even when it is unpublished. Read it as a clinician's impression, which is what it is.

The distinction changes what happens next. True narcolepsy might respond to drugs. REM deficiency needs somebody to work out why the horse is not lying down, and the answer is usually pain that makes standing up expensive, a surface he does not want to be on, or a stable where he has never felt safe enough to try.

What we say, and what it might be

Figure 6

Five readings worth swapping

What we sayWhat it might be

Unmotivated→Sleep-short

Gives up on training→Out of perseverance, not out of ability

Collapses while being groomed→Entering REM on his feet

Lies down regularly, so he's fine→May still sleep badly if the night keeps breaking

Never lies down→Or has never been seen to. Those are not the same thing

Greening and McBride open their review by pointing out that sleep is not commonly treated as a welfare factor at all, despite being something every mammal needs. They name no framework in particular, and they do not need to. Look at any welfare checklist you have used and count the lines about rest.

When to worry, and when to call the vet

What follows is a prompt, not a test. It cannot tell you anything about your horse that you do not already know; what it can do is arrange what you know into the shape a vet will find useful. Every item in it was independently associated with suspected sleep disturbance in a survey of 1,749 owners, which is a much weaker thing than a diagnostic criterion.

Two of the five questions are about what you have observed rather than what is happening. That is deliberate, and it is the honest limit of the whole field. If the answer to "how often does he lie down" is "I don't know", that is the most useful answer you can give, because it is the one with a cheap fix.

What's still open

Almost everything. Nobody knows the REM duration below which welfare is actually compromised. Nobody knows how much of the variation between horses is temperament and how much is the stable, and the authors of the Helsinki study say so. There is no figure for how long recovery takes once management improves, no non-invasive tool an ordinary owner can use at home, and no evidence yet on whether the sleep and willingness relationship holds outside a riding school, in older horses, or in the horses with clinical sleep symptoms that the study specifically excluded.

There is enough, though, to stop filing the sleep-short horse under training problem.

What I did about Giles, and what it does not prove

He is not in that stable any more. He moved to track livery in the spring, and he is out with a herd, and I now see him lie down. Flat out, in company, in the middle of the afternoon, while somebody else stands over him looking about. Every time I see it I think the same comfortable thought: they take turns.

I would like to tell you the research supports that. Having spent a fortnight in it, I have to tell you it does not.

Start with the turn-taking, because it is the loveliest part and the weakest. In the equine literature that claim rests on a single article from 1990, in a small American practice journal, with no data behind it and no retrievable text. Greening and McBride repeat it and hedge it as a supposed rota. Meanwhile the largest study ever done on how horses coordinate rest, 126 feral horses filmed by drone across 20 days and 23,716 position records, found the opposite of a rota. Within a family group horses matched each other's state 93% of the time. They go down together and they get up together. The paper does not mention sentinels once.

There is a human version of this that I cannot stop thinking about. Researchers put activity monitors on a Hadza camp in Tanzania for 20 nights. Across the whole period, the total time when everyone was asleep at once came to 18 minutes. Near-perfect cover, all night, every night, and nobody was on a rota. It was just that the old people went to bed early and the young ones stayed up.

Then the move itself. There is no published study showing that taking a horse out of a stable and putting him in a herd improves his sleep. The best-designed test of almost exactly that comparison, published last year, found lying time went up in the expected direction and could not support it statistically. Another study put the same eight horses through single boxes and group shelters: they lay down for 145 minutes a night in a box, and 69 minutes in a group shelter at the legal minimum space. In the shelters, every horse was pushed back to its feet by another horse at least once. That cannot happen in a stable. The only track-system study with recumbency data has the track slightly lower than the comparison.

What group living does have behind it is a stress measure rather than a sleep one. Horses kept in isolation show meaningfully higher faecal glucocorticoids than horses in a group. I think the move was right. I do not have evidence that it was right for the reason I have been telling people.

Here is the part that took me longest to admit. Every study in this article measures posture. Not one of them measured Giles, and not one of them used EEG. When I say I know he sleeps better now, what I actually mean is that I can see him lying down, and I could not before. I did not acquire evidence. I acquired a view.

Which is the same mistake as the shape in the shavings, made outdoors, in better light.