There is a version of sleep advice that goes like this: sleep in 90-minute blocks, set your alarm for a multiple of 90 minutes from bedtime, and you will wake up refreshed.
It is not wrong, exactly. It is just imprecise in a way that causes real confusion when someone's alarm lands at 6:30 and they wake up feeling like they were hit by a truck. Their cycles were not 90 minutes that night. They were 104, then 96, then 118.
Here is what the sleep cycle actually is, how long each one really runs, and what changes the shape of the night.
TL;DR: A sleep cycle is one pass through light sleep, deep sleep and REM, and it lasts roughly 70 to 100 minutes in the first cycle and about 90 to 120 minutes in later cycles. Most adults complete four to six cycles a night. What matters more than cycle length is the shape of the night: deep sleep is front-loaded into the first third, REM is back-loaded into the last third, and the last couple of cycles contain almost no deep sleep. That is why cutting the final 90 minutes costs proportionally more REM than cutting the first 90 minutes costs deep sleep. The 90-minute rule is a useful rough guide, not a schedule your body follows to the minute.
YouTube: Related video
What a sleep cycle actually is
Sleep is not a single state and it is not a smooth decline into unconsciousness. It is a repeating sequence with a fairly rigid order.
A healthy adult falls asleep through N1, the few minutes of drifting at the edge, then spends the bulk of the cycle in N2, drops into N3 (slow-wave or deep sleep), climbs back through N2, and finishes with a REM period before the cycle restarts.
The StatPearls summary of sleep physiology describes the sequence as N1, N2, N3, N2, REM, with each stage carrying distinct brain wave patterns, muscle tone and eye movement signatures. That order is remarkably consistent: you do not skip from N1 straight into REM as an adult, and you do not start the night in REM.
Two things that surprise people:
You spend most of the night in N2. N2 is not a minor transitional state. It is the largest single component of sleep, and it is where sleep spindles and K-complexes appear, the EEG features associated with memory consolidation.
Each stage has a different job, and the timing of that job is scheduled. N3 clusters in the first third of the night. REM gets longer in the last third. The reference texts describe this pattern directly: most slow-wave sleep occurs early, and REM episodes lengthen as the night progresses, with the first REM period sometimes lasting only one to five minutes.
How long each cycle actually lasts
The commonly cited figure is 90 minutes. The research literature gives a range, and the range shifts across the night.
| Cycle | Typical length | What dominates it |
|---|---|---|
| First cycle | 70 to 100 minutes | Largest block of N3 (slow-wave) sleep, roughly 20 to 40 minutes of it |
| Second cycle | 90 to 120 minutes | Less N3, longer first REM period |
| Third and later | 90 to 120 minutes | Mostly N2 and REM, N3 may be minimal or absent |
| Whole night | 4 to 6 cycles | About 75 to 80% NREM, 20 to 25% REM |
Harvard's sleep medicine materials put the first NREM-REM cycle at 70 to 100 minutes and later cycles at 90 to 120 minutes. The Institute of Medicine's sleep physiology chapter gives the same numbers from Carskadon and Dement: a first cycle of 70 to 100 minutes, later cycles of approximately 90 to 120 minutes.
InformedHealth expresses it slightly differently, as cycles occurring every 90 to 110 minutes, which works out to about five cycles in an eight-hour night.
Notice that the first cycle is shorter than the rest. This is the detail that breaks the alarm-multiple trick. If you set your alarm for six hours after bedtime assuming perfect 90-minute blocks, you are predicting the cycle boundaries of an average person on an average night, not your own.
Why the night changes shape
The architecture shifts because the two kinds of sleep are regulated differently.
Deep sleep is driven by sleep pressure, the accumulation of adenosine and other homeostatic factors while you are awake. That pressure is highest when you fall asleep, so the first opportunity to discharge it produces the longest slow-wave blocks. By the third cycle, most of that pressure is gone and N3 thins out or disappears entirely.
REM sleep is tied more closely to the circadian clock and to time since sleep onset. Its episodes get longer as the night goes on, and in the final one-third of the night they can last 30 to 45 minutes at a time.
That asymmetry has one very practical consequence: the last 90 minutes of sleep is not equivalent to the first 90 minutes. Truncate the start of the night and you lose deep sleep. Truncate the end and you lose REM. Both matter, and they matter for different things, which is why the "I'll just go to bed later and wake up at the same time" strategy is not neutral.
This is also the mechanism behind a pattern many people notice: after a few short nights, they feel foggy and emotionally flat rather than physically wrecked. The REM-heavy end of the night is the part that got cut.
For the broader question of how much sleep you actually need, the sleep requirements by age guide covers it. Cycle structure is a separate issue from total duration.
Why waking between cycles feels different from waking mid-cycle
Waking from N3 is disorienting. Blood pressure and heart rate take time to come back up, the brain is in high-amplitude slow-wave activity, and the result is sleep inertia: the heavy, slow, mildly stupid feeling that can last 20 to 30 minutes.
Waking from N1, N2 or the tail end of REM is much easier. Those states are closer to wakefulness in terms of arousal threshold and cerebral blood flow.
The 90-minute plan is an attempt to exploit this, and the underlying logic is sound: if you can land the alarm in light sleep rather than deep sleep, you will feel better. The problem is precision. Your cycle length varies with age, sleep debt, alcohol, temperature and the previous night, so a fixed multiple will sometimes land perfectly and sometimes land in the middle of a slow-wave block.
If you want to try it, the more robust version is to aim for a wake time and work backward, rather than the reverse: pick a consistent wake time, go to bed 7.5 to 8.5 hours before it, and let the cycles land where they land. Consistency stabilises the architecture, which makes the boundaries more predictable over time.
What changes your sleep architecture
Total sleep time and stage proportions both respond to behaviour, and not always in the way people expect.
Alcohol is the biggest single disrupter of architecture. It shortens sleep latency, which feels sedating, then suppresses REM in the first half of the night and produces a rebound of fragmented, REM-heavy, easily disrupted sleep in the second half. If you track sleep stages on a watch, the fingerprint is usually reduced REM and elevated heart rate overnight. The guide to what alcohol does to your sleep and recovery goes deeper on this.
Caffeine extends sleep latency and reduces total sleep time, and its effects can persist for six hours or more depending on when you take it.
Heat fragments the second half of the night, which is why a warm bedroom tends to cost you REM and produce more awakenings.
Training load can deepen slow-wave sleep modestly on hard days, but heavy late-evening sessions raise core temperature and heart rate at exactly the wrong moment.
Age shifts the whole distribution. Deep sleep declines steeply from the twenties onward, and older adults spend more time in lighter stages with more awakenings. That is normal, and it is one reason strict stage targets are a poor idea for anyone over 40.
What your watch is actually showing you
Apple Watch reports sleep as Awake, REM, Core and Deep. Core is N1 plus N2 collapsed into one bucket, Deep is N3, and REM is REM.
The stage-level accuracy is moderate at best. Wrist-based estimation infers stages from movement and heart rate patterns rather than measuring brain activity, so individual nights are noisy. Aggregates over weeks are more stable and more useful. The Apple Watch sleep stages guide covers what each stage means in the app, and the sleep stages accuracy piece covers where the estimates drift.
What the watch is genuinely good at, and what matters more than stage labels:
- Total sleep time and time in bed, which is the base of everything else.
- Sleep consistency, measured as the variation in your sleep and wake times.
- Overnight heart rate and HRV, which respond to alcohol, illness and heat before you notice any of it subjectively.
If your deep sleep reading swings from 40 minutes to 90 minutes and back within a week, that is more likely measurement noise than physiology. If your REM percentage trends down for three weeks while your sleep duration is unchanged, that is worth a look.
The 90-minute rule: what to keep and what to drop
Keep the idea that sleep runs in cycles with a light-sleep exit point roughly every hour and a half. It is a genuinely useful mental model, and it explains why a 20-minute nap works while a 50-minute nap leaves you groggy.
Drop the idea that you can schedule around it precisely. Cycle length varies by 20 to 30 minutes between individuals and shifts within a single night, as the Harvard and Institute of Medicine sources both describe. An alarm set to a multiple of 90 minutes is a guess with a decent prior, nothing more.
What actually improves how you feel on waking, in rough order of effect:
- A consistent wake time, seven days a week. This anchors the circadian component and makes cycle boundaries more predictable.
- Enough total time in bed that you complete five or six cycles rather than four.
- Less alcohol, which is the single fastest way to restore normal architecture.
- A cool, dark room, which protects the REM-heavy second half of the night.
- Daylight in the morning, which sets the timing of the whole system.
If you want a single metric to watch, watch consistency rather than cycle maths. Century AI folds your sleep timing, overnight heart rate and HRV into one daily score, which is useful precisely because it rewards regularity over one good night.
A realistic night, planned backwards
If you want to wake at 06:45 and feel decent, the practical version looks like this:
- Lights out around 22:45 to 23:15. That gives seven and a half to eight hours in bed, which after normal sleep latency is roughly seven hours of sleep and five complete cycles.
- Expect the first cycle to end around 00:00 to 00:30. This is your deepest sleep. If you wake here, you will feel awful, and that is normal.
- Expect the last cycle to be REM-heavy. If you wake naturally at 06:30 instead of 06:45, get up. You probably finished a cycle.
- Do not chase the alarm multiple if it means cutting sleep. Eight hours plus a slightly awkward wake time beats six hours in perfect 90-minute blocks.
And if you wake up mid-cycle anyway, which you will, the recovery move is the same as always: light, water, movement, and about 20 minutes of patience for the inertia to clear. There is a reason the sleep inertia guide exists.
What not to do with this information
The most common failure mode with sleep cycle knowledge is turning it into another thing to optimise. Counting cycles, calculating bedtime from a stopwatch, and worrying about whether your deep sleep hit 18% or 20% is a fast route to the sleep-tracking anxiety that has its own name.
The useful part of cycle science is structural, not numerical. Deep sleep comes early, REM comes late, cycles get longer as the night goes on, and consistency is what keeps the pattern stable. Build your evening around those four facts and ignore the arithmetic.
Century AI helps you understand your body with a daily health score, recovery score, and sleep insights — using the watch you already wear.
