What a sleep cycle actually is, and why 90 minutes is an average, not a rule

What is actually inside a cycle

Sleep is not one state you drop into and climb out of. It is a repeating progression through four scored stages, and a "cycle" is one full pass through that progression. The standard order is N1 into N2, down into N3, back up through N2, and then into REM — after which the whole thing starts again.

The stages are not equal partners. This is the part that surprised me most when I first looked at the normative figures, because the popular framing treats deep sleep and REM as the main event when in fact the bulk of your night is spent in ordinary light sleep:

Roughly half of your night, then, is N2 — the stage nobody markets. Deep sleep, the one every wearable puts a big number next to, is under a fifth of it.

The shape of the night changes as it goes

Here is the thing that makes the "count 90-minute blocks" model leakier than it looks: the cycles are not interchangeable units. A cycle at 11pm and a cycle at 5am contain very different things.

Early cycles are dominated by deep slow-wave sleep, with only brief REM episodes attached to the end. By the second half of the night, deep sleep has largely dropped out of the cycle altogether and REM has expanded to fill the space. The first REM episode of the night might last about 10 minutes; a late one can stretch past 30.

The mechanism is straightforward once you see it. Slow-wave sleep is driven by accumulated sleep pressure, which is at its maximum the moment you fall asleep and discharges steadily overnight. Your body spends the pressure early, on the stage that needs it most, and then has room for REM later.

Two practical consequences fall out of this. First, cutting your night short does not remove a proportional slice of every stage — it disproportionately removes REM, because REM is stacked at the end. Second, going to bed much later than usual disproportionately costs you deep sleep, because you are compressing the front of the night.

Where the 90-minute number came from

The figure traces back to the foundational descriptive work on normal human sleep, where Carskadon and Dement characterised the adult cycle and put individual cycle length at approximately 70 to 120 minutes. Ninety is the round number sitting in the middle of that.

What happened next is the part worth noticing. The range got dropped and the midpoint survived. Ninety travelled out of the sleep literature into productivity blogs, then into alarm clock apps, then into every sleep calculator on the internet — each hop quietly upgrading it from "central tendency of a wide distribution" to "the length of a sleep cycle."

How wide the real spread is

Wide enough that I had to think about whether the calculator was defensible at all.

• Within one person, across one night. Cycle length is not fixed even for you, even tonight. The first cycle is often the shortest, commonly 70 to 100 minutes, with later cycles running longer at 90 to 120.
• Between people. The classic 70–120 minute range describes healthy adults generally. Your personal centre could sit anywhere inside it.
• In large modern datasets, higher than the textbook. A retrospective analysis of 16,441 nights from 573 subjects found a median ultradian cycle length of 110 minutes, with 75% of cycles falling between 95 and 130 minutes. That is meaningfully above 90.
• Cycle content drives cycle duration. Length is not an independent clock ticking underneath. What the cycle contains — how much slow-wave sleep, how much REM — influences how long it runs, which is why the shape of the night and the length of its cycles change together.

Sit with that last dataset for a second, because it is the one that should give any calculator author pause. If the median is nearer 110 than 90, then five "cycles" for a median sleeper is 9 hours 10 minutes, not 7 hours 30. The arithmetic every sleep calculator performs, including mine, could be a full cycle out.

So why did I keep 90 in the calculator?

Because a schedule you will actually follow beats a precise number you will not, and because the alternative was worse.

I considered three options while building it. I could drop cycles entirely and just recommend a duration — honest, but it throws away the one framing that gets people to pick a bedtime and stick to it. I could ask everyone to measure their own cycle length, which requires equipment almost nobody has and patience almost nobody has either. Or I could use the conventional average, be loud about what it is, and let people change it.

I went with the third, which is why there is a slider on the calculator that moves cycle length from 70 to 120 minutes, and why the assumptions are printed on the homepage rather than buried. If you have tracker data suggesting your cycles run long, move the slider to 110 and the whole schedule shifts with it.

What I would not do is what the rest of the category does, which is present 90 minutes as settled fact and the resulting bedtime as an optimum. The number is a convention. It produces a reasonable schedule. It does not produce a guarantee about which stage you will be in when the alarm goes off, and I have written separately about exactly how much confidence the output deserves.

References

• Patel AK, Reddy V, Shumway KR, Araujo JF. Physiology, Sleep Stages. StatPearls, NCBI Bookshelf. Source for the stage proportions (N1 3–8%, N2 45–55%, N3 15–20%, REM 20–25%).
• Ultradian sleep cycles: frequency, duration, and associations with individual and environmental factors — a retrospective study. The 16,441-night dataset giving a median cycle length of 110 minutes.
• Feinberg I, Floyd TC. Sleep cycle content and sleep cycle duration. On cycle content influencing cycle length.
• Sleep Foundation. Stages of Sleep: What Happens in a Normal Sleep Cycle? Accessible summary of the 70–120 minute range and the front-loading of deep sleep.
• American Academy of Sleep Medicine. Source of the adult 7-or-more-hours recommendation referenced throughout this site.
• Accuracy of 11 wearable, nearable and airable consumer sleep trackers: prospective multicentre validation study. JMIR mHealth uHealth, 2023. Source for the staging-accuracy figure cited in the first-hand section.

What this looked like on me

For six weeks in early 2026 I kept a record of when I got into bed, roughly when I thought I had fallen asleep, and when I woke. The instrument was the stopwatch app on a Samsung phone. That is not a sleep tracker and I want to be precise about what it means: I have no stage data at all. Nothing in this article about deep sleep being front-loaded, or about REM lengthening toward morning, is something I observed. I read it.

Oddly, I think that is the more defensible position. A wearable would have handed me a confident-looking breakdown of light, deep and REM, and consumer devices are only 60–85% accurate at staging when checked against polysomnography. Two timestamps and no algorithm at least leaves me clear about what I do not know.

What matched: I woke naturally at roughly the same point most mornings, without an alarm and within a fairly narrow window. That is consistent with sleep being organised into repeating units rather than simply running until something interrupts it, which is the one structural claim in this article my own data touches at all.

What did not match: the arithmetic. Working backwards from when I fell asleep to when I woke naturally, the intervals came out consistently shorter than a multiple of ninety minutes would predict. Whatever my cycle length is, ninety appears not to be it — and that is why I moved the slider on my own calculator off the default, which is a slightly absurd thing to have to do to a tool you built.

The honest caveat is that this is an inference drawn from two timestamps, not a measurement. A stopwatch cannot see a sleep cycle. I also cannot tell you whether my first cycle differs from my fourth, which the literature says it should. All I can say is that the default did not fit, which is the argument this article was already making.

How this was checked

A general policy is easy to write and hard to hold anyone to, so here is what was actually done for this page — including the parts that came up short.

• Stage proportions (N1 3–8%, N2 45–55%, N3 15–20%, REM 20–25%) are taken from StatPearls, a tertiary clinical reference, and cross-checked against my own physiology course material.
• The 70–120 minute range is attributed to Carskadon and Dement's descriptive work on normal human sleep. I reached it through secondary citation rather than the original chapter. Flagging that, because the article leans on this figure heavily.
• The 110-minute median comes from the retrospective analysis of 16,441 nights. I read the methods specifically to check how the authors defined a “cycle”, because different definitions produce different numbers and that is exactly where this sort of figure goes wrong.
• The hypnogram is drawn to show shape only. It is not measured data from any individual, and the caption says so on the page rather than only here.
• What I could not establish: how much of the 70–120 minute spread is variation between people versus variation within one person across a night. Both are real; their relative size I do not know, and the article does not claim to.

Every article here is researched against the sources listed above, drafted with AI assistance, and then reviewed and approved by me before publication. AI supplies drafting speed, not judgement: it does not choose the topics, decide what is true, or generate a single citation. No reference on this site is one I have not opened. The full policy, including why AI is used at all, is here.

What this page does not cover

Stated plainly so you are not left hunting. Where somebody answers one of these better than I can, that is where the link goes.

• Sleep architecture in infants and children. It differs substantially from the adult pattern described here, and none of these proportions transfer. American Academy of Sleep Medicine →
• Architecture in sleep disorders. Obstructive sleep apnea, narcolepsy and others reshape the night in ways this page does not describe. That is a clinical assessment, not a reading task.
• How to deliberately get more deep sleep. A fair question that deserves its own page rather than a paragraph here.
• Dreams, and what they mean. Outside what I can source to a standard I would be comfortable publishing.

Related reading

• How much sleep do I actually need — the guideline numbers by age, and what "need" means.
• Are sleep calculators accurate? — an honest audit of the tool I built.
• The sleep calculator — with the cycle-length slider this article argues for.