The Stages of Sleep: What Happens in Light Sleep, Deep Sleep, and REM

Short Answer

Sleep isn't a single flat state. It moves through a repeating architecture: NREM sleep first, then REM sleep, again and again across the night. NREM is split into N1 — the light drift-off between waking and sleep, N2 — the steadier "you're asleep now" stage that takes up a large part of the night, and N3 — deep, slow-wave sleep. REM is the more brain-active stage when most vivid dreaming happens. According to NHLBI, one sleep cycle restarts about every 80–100 minutes, and most people move through four to six cycles per night. (NHLBI)

Each stage is doing different work in your body. N3 is the deepest NREM stage; you usually get more of it early in the night, when sleep pressure is high, and this is one reason a short, broken night can leave you feeling physically unrefreshed even if you technically "slept." REM tends to build later, toward morning, when the brain is active, dreaming is more likely, and memory and emotion-related processing are in play. Sleep also changes with age: slow-wave sleep peaks in childhood, drops sharply in the teenage years, and tends to keep decreasing through adulthood. (NHLBI)

N2 deserves attention too. It's not "just light sleep." Sleep spindles — brief bursts of brain activity that are a hallmark of N2 — appear to support learning and memory. As one 2025 study notes, sleep spindles "are involved in memory consolidation, promote memory and learning, and are significantly associated with physiological aging and cognitive decline" (Scientific Reports, 2025).

The important caveat: your watch is estimating sleep stages, not measuring them the way a sleep lab does. Lab sleep staging uses sensors that record brain activity and eye movement; consumer wearables usually infer stages from signals like movement, heart rate, and related patterns. That inference can be useful for trends, but it is not a diagnosis. N1 is especially hard to classify because it sits in the borderland between wake and sleep; as a 2026 paper puts it, "automated sleep staging remains challenging due to the transitional nature of certain sleep stages, particularly N1" (Biosensors, 2026).

The stages of sleep at a glance

The two big families: NREM and REM sleep

At the highest level, your night is built from two very different kinds of sleep: NREM (non-rapid-eye-movement) sleep and REM (rapid-eye-movement) sleep. NREM is the quieter, more physically restorative side. It has three depths — N1, N2, and N3 — and your brain activity changes as you move from the light edge of sleep into deep, slow-wave sleep. REM is different. Your brain becomes much more active, closer to a waking pattern; your eyes move quickly under closed lids; dreaming usually shows up here; and your major muscles normally go limp so you do not act out those dreams. (NHLBI)

That split matters because sleep is not one uniform "off" state. In sleep studies, stages are classified using signals such as brain activity and eye movement, which means NREM and REM are not just labels for how sleep feels — they are different physiological states. They also seem to support different jobs. Research consistently treats them as separate axes of sleep quality: a 2026 review of sleep and cognition, for instance, links "slow-wave sleep (SWS) to exercise-induced neuroplasticity and sleep spindles to memory consolidation" (Frontiers in Neurology, 2026), while describing REM as part of a distinct cognitive pathway.

So when you look at sleep stages, you are not looking at a tidy staircase from "light" to "deep" and then "dreaming." You are looking at several body-and-brain processes stacked together across the night: lighter NREM sleep that helps you transition and stabilize, deeper NREM sleep where slow-wave activity becomes prominent, and REM sleep where the brain is active while the body is kept still. Understanding that architecture makes wearable sleep-stage charts easier to read — and easier not to overread.

N1 and N2 — light sleep (and why N2 matters more than it sounds)

N1 is the doorway into sleep. You're drifting, but not far gone: your body is starting to downshift, your awareness is loosening, and a sound, a thought, or a small movement can pull you right back out. Many people don't even feel like they were asleep during N1. It usually lasts only a few minutes, and that in-between quality is exactly what makes it hard for sleep-tracking algorithms to label cleanly. In lab research, N1 is described as a transitional stage between wakefulness and sleep, and automated staging still struggles with it: "automated sleep staging remains challenging due to the transitional nature of certain sleep stages, particularly N1" (Biosensors, 2026). That's one reason your wearable's sleep-stage chart should be read as an estimate, not a direct measurement of your brain waves.

N2 sounds unglamorous because it still gets grouped under "light sleep." But it's not filler. It's usually the largest part of the night, and it's where your sleeping brain starts doing more structured work. One of N2's signature features is the sleep spindle — a short burst of rhythmic brain activity. As a 2025 study defines them, spindles are "characteristic electroencephalographic hallmarks of nonrapid eye movement sleep (NREM), typically defined as oscillatory activity in the sigma frequency range (11-16 Hz) lasting 0.5-3.0 s" (Scientific Reports, 2025). Those bursts aren't random static. They're part of the way your brain protects sleep, filters sensory input, and helps stabilize what you learned while you were awake. The same study puts it plainly: sleep spindles "are involved in memory consolidation, promote memory and learning, and are significantly associated with physiological aging and cognitive decline" (Scientific Reports, 2025). So the "light" part of your night is still doing brain work — helping yesterday's practice, reading, conversations, and problem-solving become less fragile.

In adults, N2 commonly makes up about 45% of total sleep, and it is often described as roughly the largest single stage of the night, which is why small night-to-night shifts in this stage can make a sleep chart look dramatic even when your overall sleep is fairly typical. NHLBI's patient guide describes stage 2 non-REM as the stage where "you are asleep," and clinical overviews place it at about 45% of total sleep. (NCBI Bookshelf)

N3 — deep (slow-wave) sleep and physical repair

N3 is deep sleep, also called slow-wave sleep (SWS) because your brain shifts into large, slow delta-wave activity. This is the heaviest part of non-REM sleep: your arousal threshold is high, so if someone wakes you here, you may come up slowly, foggy and disoriented rather than instantly alert. That grogginess is sleep inertia — your brain is being pulled out of a low-frequency, high-recovery state before it has had time to "boot up." N3 also clusters toward the first half of the night, which is one reason the early hours of sleep can feel so physically restorative. NHLBI describes stage 3 as deep or slow-wave sleep and notes that you usually spend more time in it early in the night; NCBI Bookshelf similarly describes N3 as the deepest non-REM stage and the hardest stage to wake from. (NHLBI)

Functionally, N3 is where a lot of the body's overnight maintenance happens. Heart rate and breathing tend to slow, muscles are relaxed, and the body leans into repair: tissue recovery, immune support, and the kind of "restore the system" work you do not feel happening in the moment but notice the next day as steadier energy. The brain is also using this state for cleanup. As a 2026 review describes, the glymphatic system is "a sleep-dominant clearance mechanism" (Frontiers in Neurology, 2026) — the brain's overnight "rinse cycle," most active during slow-wave sleep, when cerebrospinal fluid movement helps clear metabolic waste. The same review connects SWS with cognition, noting that "SWS duration correlates with inhibitory control and working memory" (Frontiers in Neurology, 2026); in plain English, deep sleep is not just "body sleep." It also helps the brain keep tomorrow's attention, self-control, and working memory online.

Deep sleep is a meaningful marker of sleep quality more broadly, but it is not a magic score to chase in isolation. In a large 2026 wearable study of 52,952 All of Us participants, researchers found that longer deep and REM sleep were inversely associated with incident atrial fibrillation, while more light sleep and long non-main sleep were linked with higher risk. That does not mean your watch can diagnose your heart health from one night of sleep stages. It does mean that, across many people, deeper and more consolidated sleep architecture tended to travel with better cardiovascular patterns. (JACC: Advances, 2026)

Deep sleep is commonly cited as about 25% of sleep in adults, especially in younger adults, but that share is not fixed. It changes with age, sleep pressure, illness, alcohol, medications, stress, and how fragmented the night is. NHLBI notes that slow-wave sleep peaks in early childhood, drops sharply during the teenage years, and continues to decrease through adulthood; NCBI Bookshelf also describes N3 as decreasing with age. So if your wearable shows less deep sleep than you expected, read it as a pattern to investigate, not a verdict on whether you "recovered enough." (NHLBI)

REM sleep — dreaming, memory, and emotion

REM (rapid-eye-movement) sleep is the stage most people mean when they talk about dreaming. Your brain is switched on — its activity looks closer to waking than to deep sleep — while most voluntary muscles normally go limp. That temporary "off switch" is protective: it helps keep dream content from turning into real movement. REM also tends to become more prominent later in the night, which is why the dream you remember may be the one that happened close to morning. (NHLBI)

REM seems to specialize less in physical repair and more in what your brain does with experience. It helps the mind test connections: what belongs together, what can be updated, what a memory feels like once the emotional charge has been worked through. The 2026 cognition review notes that "REM sleep is linked to cognitive flexibility" (Frontiers in Neurology, 2026), and REM is widely associated with memory integration and emotional processing.

That does not make REM "just psychological." In the large 2026 All of Us wearable cohort, researchers analyzed sleep data from 52,952 participants and found that deep and REM sleep duration were both "inversely associated with AF" (atrial fibrillation) (JACC: Advances, 2026). In plain language: people with less time in these stages tended to have higher incident AF risk in that observational dataset. That does not prove that forcing more REM prevents arrhythmia, but it is a useful reminder that sleep architecture tracks with physical health, not only mood or dream recall.

REM typically accounts for about 20% to 25% of an adult's night — roughly a quarter — but it is not a number you need to chase minute by minute. One-line definition: REM sleep is the active, often dream-filled stage of sleep when the brain looks wake-like while most voluntary muscles are temporarily switched off. (NCBI Bookshelf)

Sleep cycles: how the stages repeat across the night

You don't spend one neat block in each sleep stage and then "graduate" to the next. Your brain and body keep looping through them in cycles. A typical cycle moves through NREM sleep — from light N1 into steadier N2 and, when it happens, deep N3 — and then into REM. Adult sleep-analysis literature commonly describes these NREM/REM cycles as lasting about 90–110 minutes; NHLBI's patient guidance gives a very similar practical range, noting that the cycle restarts every 80–100 minutes. So the useful takeaway is not an exact clock time, but the rhythm: across a full night, most people move through about four to six cycles. (NHLBI)

The important part is that each cycle is built differently as the night goes on:

• Early cycles are deep-sleep-heavy. This is when your brain is under the strongest pressure to recover from the day. More of your N3, or deep slow-wave sleep, tends to happen in the first part of the night. (NHLBI)

• Later cycles are REM-heavy. As morning gets closer, REM takes up more space in the cycle. That is why vivid dreaming is more likely in the back half of the night, and why sleeping only the first few hours is not the same as getting a whole night. (NHLBI)

This is why when you sleep, and whether you sleep long enough to complete several cycles, changes the kind of sleep you actually get. A 7–9-hour night gives your body room to move through repeated cycles; cutting the night short tends to shave off the REM-rich final stretch, while frequent awakenings can break up the architecture that lets each stage do its job. (NHLBI)

That architecture matters in research, too. Total sleep time can tell you how long you were asleep, but it cannot show whether that sleep was mostly light, deep, REM, consolidated, or fragmented. That is why researchers increasingly look at the pattern of stages, not just the total minutes. As one 2026 wearable study notes, many studies are "limited by the use of self-reported data or absence of information regarding sleep architecture" (JACC: Advances, 2026).

How sleep stages change with age

Sleep architecture is not fixed for life. It changes as your brain and body change. Newborns spend more of the night in REM sleep, while the slow-wave patterns that later show up as deep N3 sleep are still maturing. In infancy, that slow activity is not just "rest" on an EEG — it is part of how the cortex develops. A 2026 longitudinal EEG study of 3- to 6-month-old infants notes that "slow wave activity (SWA) is a key marker of cortical maturation and experience-dependent plasticity" (npj Biological Timing and Sleep, 2026), and found measurable regional changes across those first months of life.

Later, the curve bends the other way. Slow-wave sleep peaks in early childhood, drops sharply during the teenage years, and keeps declining through adulthood; in older age, sleep often becomes shorter, lighter, and more broken by nighttime awakenings. That does not mean older adults "need less sleep." They generally still need about the same full night as other adults — roughly 7 to 9 hours — but the shape of that night changes: less deep sleep, more light sleep, more waking between cycles. (NHLBI)

N2 sleep changes too. Sleep spindles — those brief bursts of brain activity tied to memory and sleep stability — tend to shift with aging, and research links spindle patterns with cognition. Recall that spindles are "significantly associated with physiological aging and cognitive decline" (Scientific Reports, 2025). So if your wearable shows a smaller deep-sleep share than it used to, or less than a younger friend's, age may be part of the explanation. It is not automatically a red flag. The better question is whether your sleep is changing for you: more awakenings, less total sleep, worse recovery, daytime sleepiness, or a sudden drop from your usual pattern.

Why wearables estimate stages imperfectly — the Welltory angle

Here's the honest part: your sleep-stage chart is an estimate, not a direct recording of your sleeping brain. The gold-standard test is polysomnography in a sleep lab. It records brain waves and usually combines them with eye-movement, muscle, heart-rate, breathing, and oxygen signals, which is why clinicians can use it to diagnose sleep disorders and score sleep stages much more directly than a consumer device can. (NHLBI)

Your watch or ring does something different. It looks at signals it can capture from the outside — movement, heart-rate patterns, photoplethysmography, sometimes breathing or temperature — and then an algorithm infers which stage you were probably in. That can work reasonably well for telling "asleep" from "awake," but the body does not flip cleanly from one stage to another. The edges are blurry. N1, the light transitional stage between wakefulness and sleep, is especially hard even for automated systems using EEG data: "automated sleep staging remains challenging due to the transitional nature of certain sleep stages, particularly N1" (Biosensors, 2026).

That's why the same device can be useful and imperfect at the same time. In one validation study, a Garmin Vivosmart 4 detected sleep with high accuracy — 0.90, with sensitivity of 0.98 — but its agreement for specific sleep-stage detection was much weaker, with Cohen's kappa of 0.20. In another study comparing five commercial sleep-tracking devices with polysomnography, total sleep time was often reasonably close, while stage accuracy varied by sleep stage, device, time spent in that stage, and the individual person wearing it. (PLOS ONE, 2020)

So what should you do with the "deep sleep: 1h 12m" number in your app? Treat it as a trend signal, not a lab report. It can help you notice your own patterns over weeks: does deep sleep tend to look worse after alcohol, jet lag, late workouts, illness, or a high-stress stretch? Does REM drop when your schedule gets chaotic? That pattern may be useful. But a single night's stage breakdown is not a diagnosis, and it is not a fair way to compare one brand against another, because each algorithm may draw the stage boundaries differently.

The best use is context. Look at sleep stages beside how you feel, your bedtime consistency, your resting heart rate, your HRV, and your recovery signals. If your tracker keeps showing poor sleep and you also have loud snoring, gasping or breathing pauses, severe daytime sleepiness, morning headaches, or trouble staying awake, that is a reason to talk with a clinician — not to self-diagnose from the app. For a full breakdown of how accurate sleep trackers actually are and how to read their scores, see our sleep tracking guide. And because stage estimates lean heavily on heart-rate patterns, tracking your HRV can give useful, complementary context on how recovered your body actually is.

How we made it

Made with AI tools, then edited, fact-checked, and medically reviewed by the Welltory team.

Discounts for blog readers: up to 36% off

See what affects your energy, stress, sleep, and daily state with Welltory

Explore your data