← Longevity & Supplement Guides
How REM sleep was discovered — and how it became a number on your wrist
The night the pens went wild
In December 1951, a thirty-year-old graduate student hauled an obsolete brain-wave machine out of a basement at the University of Chicago and wired it up to his eight-year-old son.
Eugene Aserinsky had reason to be careful with borrowed equipment. He was living with his family in converted army barracks on campus, broke enough that he would eventually borrow money from his own dissertation adviser, and he got through the long overnight sessions on pretzels and coffee. The machine was an Offner Dynograph. The room was on loan from the physiology department in Abbott Hall. The subject was Armond, who was asleep.
Then the pens started to swing. Great sweeping bursts of ink, the kind of deflection you get from someone thrashing around — except the boy was lying perfectly still.
Aserinsky's first hypothesis was the obvious one: the machine was broken. That is a sensible instinct, and it is also the reason most anomalies quietly die. He went and looked at his son instead. Under the closed lids, the eyes were moving — fast, jerky, weirdly coordinated, nothing like the slow drift of someone settling down for the night.
That is the whole discovery, in the sense that everything after it was work. Aserinsky wrote his own account of it forty-five years later, in the Journal of the History of the Neurosciences, and it is worth reading for how unglamorous he allows it to be.
Before that night, sleep was nothing happening
To feel why this mattered, you have to appreciate how boring sleep was supposed to be.
The dominant view was passive. The brain wound down, activity subsided, and the night was the trough between two days. Sleep was defined by what it lacked. There was no particular reason to go looking for structure inside it, in the same way there is no reason to go looking for furniture in an empty room.
The irony is that the man who had done more than anyone alive to make sleep a serious science held roughly that view himself. Nathaniel Kleitman published Sleep and Wakefulness in 1939 and more or less founded the field in English. He was not a cautious man about it, either. In the summer of 1938 he and a graduate student, Bruce Richardson, moved into Mammoth Cave in Kentucky, about 140 feet below the surface, and lived there for 32 days on a 28-hour day — six long days to the calendar week, temperature readings every two hours awake and every four asleep. Richardson's body largely took to the new schedule. Kleitman's refused, which he put down to being the older of the two.
That is a man willing to spend a month underground to find out whether the body keeps its own time — a question we now file under circadian rhythm. And even he thought of the night itself as one long uniform state.
The dud assignment
Aserinsky did not set out to find any of this. He was handed the kind of project an adviser hands out when he mainly wants a student occupied: eye blinks, and then the eyelid movements of infants. Watch babies, count things, come back in a year.
He found the babies unrewarding and drifted toward adults, and toward the equipment nobody was using. The field's habit of arriving through a side door is one of its most consistent features — the discovery that mattered came out of a project designed to be safe.
Two pages in Science
By spring 1952 he had the same pattern in other people. That autumn he ran more than fifty overnight sessions with better recording gear. On 4 September 1953, he and Kleitman published two pages in Science.
The paper is short and almost aggressively unexcited. Twenty normal adults. Recurring periods of rapid, jerky eye movement arriving roughly every ninety minutes, alongside low-voltage irregular brain activity and a lift in breathing and heart rate — the first episode brief, the last one long.
Then came the move that turned an observation into a discovery. They woke people up.
Of twenty-seven awakenings during the eye-movement periods, twenty produced detailed, visual dream narratives. Of twenty-three awakenings taken outside them, nineteen produced nothing whatsoever.
That asymmetry is the paper. It is also, if you want a lesson about measurement rather than about sleep, the entire trick: they did not merely record a signal, they asked the signal what it was evidence for.
Somebody had seen it before. That is not the same as finding it.
Now the myth-correcting part, which is usually the best part of any history.
The eye movements were not first noticed in Chicago. In 1926, two Russian researchers, Denisova and Figurin, described cyclical bursts in sleeping infants: unsettled breathing, more body movement, a faster heart rate, and the eyes moving beneath half-closed lids, recurring about every fifty minutes. They were looking straight at the right thing, twenty-seven years early.
What they did not have was the electroencephalogram — Hans Berger's human EEG arrived in 1929 — or adult subjects, or, decisively, any link to dreaming. A 2024 review in SLEEP works through the credit question and lands somewhere useful: partial observation is not discovery. Seeing something is cheap. Following it through is expensive.
That is worth remembering the next time a wellness product announces it has "discovered" a signal in your data. Signals are everywhere. Knowing what one is evidence for is the rare and costly part.
The man who found it left the field
Aserinsky finished his doctorate in 1953 and left. He went to the University of Washington to study the electrical effects of salmon, took a faculty post at Jefferson Medical College in Philadelphia, and only came back to sleep research in 1963. He was passed over for a department chairmanship in 1976, moved to Marshall University in West Virginia, and retired in 1987. Kleitman was reportedly still opening letters to him with "Dear Aserinsky" a decade after they had jointly split the night in half.
Meanwhile William Dement, a medical student who joined the lab in 1952, gave the phenomenon the name that stuck — rapid eye movement, REM — and carried the work into the decades that made it famous. Credit in science does not distribute itself by who was there first. It distributes itself by who stays.
Aserinsky died in July 1998, at 77, in a car collision in Carlsbad, California. The autopsy was inconclusive and it was considered possible that he had fallen asleep at the wheel. The record does not settle it, and I am not going to improve on the record.
The committee that turned a river into stairs
Here is where the story stops being about people and starts being about the screen you looked at this morning.
By the late 1960s everyone was recording sleep and nobody was recording it the same way. In 1968 a committee led by Allan Rechtschaffen and Anthony Kales published a manual through the US Government Printing Office: standardised terminology, standardised technique, and a scoring system that the whole field could share. It worked. For roughly forty years it was the standard.
Two of its decisions now govern how you see your own night.
First, the epoch. The night was cut into thirty-second slices, and every slice was given exactly one label. Second, the ladder: stages 1, 2, 3, 4, and REM.
Both choices were reasonable and both are administrative. Physiology does not change state on the half-minute, and a night is not really composed of five discrete conditions with clean walls between them. A thirty-second epoch is a rounding rule invented so that two laboratories could compare notes. It is the reason your night shows up as a staircase rather than as a river.
The American Academy of Sleep Medicine revisited the rules in 2007. The thirty-second epoch survived. Stages 3 and 4 were merged into a single N3, on the grounds that the boundary between them had never been worth what it cost to score. Even the standard-setters eventually conceded that one of the steps was a fiction. If you want the modern ladder in plain language, we wrote it up in sleep stages explained.
ONE NIGHT, THREE NOTATIONS
What each step of standardisation gave up
Even the experts disagree about a fifth of the night
The rulebook did not abolish judgement; it just wrote the judgement down. We know roughly how much judgement is left, because the American Academy of Sleep Medicine has measured it.
Rosenberg and Van Hout reported the results in 2013: more than 2,500 scorers, nine clinical recordings, 18,000 epochs, over three million individual scoring decisions.
Average agreement came out at 82.6%. Agreement was highest for REM. It was 67.4% for N3 and 63.0% for N1 — which happen to be the two stages people most want to be told about. Scorers struggled exactly where you would predict: the last epoch of wake before sleep begins, the first epoch of N2 after N1, the first epoch of REM after N2. The edges.
So the reference standard — trained humans, a full electrode montage, one shared rulebook — disagrees with itself about something like one epoch in six. That is not a scandal. It is what happens when you draw hard lines on a continuous process. But it is the number to hold in mind before anyone tells you their algorithm is accurate.
THE REFERENCE STANDARD, MEASURED
Inter-scorer agreement, AASM programme (Rosenberg & Van Hout, 2013)
Where the disagreement lives
The study reports that agreement was highest for stage R, with N2 and wake close behind, and that the hardest calls were transitions: the final epoch of wake before sleep onset, the first epoch of N2 following N1, and the first epoch of R following N2. Separating N2 from N3 was singled out as particularly difficult. In other words, the disagreement is concentrated at the boundaries — precisely the places a stage percentage is built out of.
How the number got onto your wrist
Your wrist device is not doing any of the above. There are no electrodes on your scalp, no leads at the corners of your eyes, no sensor under your chin. There is movement, a pulse, and a model trained to guess what a laboratory would have written down.
That guess is genuinely impressive as engineering, and it is much better at some questions than others. A 2023 multicentre validation in JMIR mHealth and uHealth put eleven consumer sleep trackers against laboratory polysomnography — 75 participants, 349,114 scored epochs. For four-class staging (wake, light, deep, REM), the macro F1 score across devices ran from 0.26 at the bottom to 0.69 at the top. The strongest REM score was around 0.76; the strongest deep-sleep score was around 0.59.
Read that range again, because the spread between devices is the finding. If the label a given half-minute receives depends this heavily on which company wrote the algorithm, then "I got 48 minutes of deep sleep" is partly a sentence about a product. We went through the general version of this problem in what your wearable can and can't measure, and the specific version in sleep beyond a single number.
What the story is actually evidence for
None of this makes sleep architecture fake. It is one of the more solid things in physiology. REM and slow-wave sleep are distinguishable states with different brain activity, different muscle tone, different hormonal weather — the reason we could write a whole post on deep sleep and growth hormone at all.
What the history does not license is reading a nightly percentage as a personal score. Three reasons to hold those bars loosely.
The proportions move with age on their own. Ohayon and colleagues pooled 65 studies and 3,577 healthy people aged 5 to 102 in 2004. Slow-wave sleep declines steadily across adulthood; the lighter stages take up more of the night; REM's share drifts down more gently. Less deep sleep than five years ago may simply be a birthday rather than a verdict on your habits.
The measurement carries a margin at every step. A committee chose the half-minute. Expert scorers disagree at the boundaries. Your device infers those boundaries from motion and pulse. Three approximations, stacked, and only the last one gets rendered as a tidy percentage with one decimal place.
The outcome research leans on other things. When large studies find sleep associated with how people age, the variables carrying most of the weight tend to be duration and regularity, not stage ratios — which we looked at in sleep regularity vs duration. Those also happen to be the two things a wrist device measures comparatively well. The number you can trust and the number that matters are, conveniently, the same number.
The bottom line
A staircase on your phone is the descendant of an accident: a broke graduate student, an unwanted machine, a sleeping eight-year-old, and a committee fifteen years later that needed two laboratories to be able to compare notes. It is a good abstraction with a real history and a known margin, and it was never designed to be a nightly grade.
So here is the one thing to try this week. Cover the stage bars with your thumb and look only at two numbers: when you went to bed, and how long you slept. Judge the week by how tightly those cluster, not by last night's REM share. Then check the stage graph once, on Sunday, as a trend — that is roughly the resolution the underlying measurement actually supports. And if a number ever contradicts how you plainly feel, the number is the thing on probation. Use numbers to correct fantasy, not to replace experience.
Aserinsky went looking at his son instead of trusting the instrument. Seventy-five years on, that is still the move.
