Sleep stages explained: what the lab measures and what your tracker estimates
A sleep lab reads brain waves. A ring or watch reads pulse, movement and temperature and makes an educated guess. Here is what each stage is and how good the guesses are.
Every morning a sleep app splits the night into coloured blocks: awake, light, deep and REM. Those labels come from sleep medicine, where stages are scored from brain activity. Consumer devices do not measure brain activity. They infer stages from other signals, and the research shows they are much better at telling sleep from wake than at telling one stage from another.
The stages, as a sleep lab defines them
In a sleep laboratory, polysomnography (PSG) records brain waves (EEG), eye movements, muscle tone, breathing and oxygen. Each 30-second epoch is scored by American Academy of Sleep Medicine rules into one of five states.
| Lab stage | Wearable label | What defines it |
|---|---|---|
| Wake | Wake | Alert or drowsy brain waves, eye blinks |
| N1 | Light | Transition from wake; slowing brain waves; easily woken |
| N2 | Light | Sleep spindles and K-complexes on EEG; most of the night is spent here |
| N3 | Deep | Slow, high-amplitude delta waves; hardest to wake from; concentrated in the first half of the night |
| REM | REM | Rapid eye movements, near-absent muscle tone, brain activity close to waking; longer periods towards morning |
Most wearables merge N1 and N2 into "light". Oura, WHOOP and Garmin all report light, deep and REM. The cycle from light to deep to REM repeats several times a night, which is why the deep blocks cluster early and the REM blocks cluster late.
How much of each stage is normal
A sleep laboratory study by the US Naval Health Research Center, published in Sleep in 2021, gives a clean reference for healthy young adults. Thirty-four people averaging 28 years old slept three nights under PSG. For the 49 nights paired with one device, PSG recorded an average of 425 minutes of sleep: 237 minutes of light sleep (about 56 percent), 81 minutes of deep sleep (about 19 percent) and 107 minutes of REM (about 25 percent). Other device subsets in the same study showed 63 to 67 minutes of deep sleep, which gives a sense of how much deep sleep varies even among similar people.
Age changes the mix. A 2004 meta-analysis of 65 studies and 3,577 people aged 5 to 102 found that in adults, total sleep time, sleep efficiency, the percentage of slow-wave (deep) sleep and the percentage of REM all decreased with age, while light sleep and time awake after sleep onset increased. A larger 2019 meta-analysis of 169 studies and 5,273 healthy adults scored under current rules found that each decade of age cut total sleep by 10.1 minutes and sleep efficiency by 2.1 percent and added 9.7 minutes of wake after sleep onset; the changes in the percentages of N2, N3 and REM per decade were not statistically significant. Sex matters too: in the Sleep Heart Health Study of 2,685 adults aged 37 to 92, women had on average 106 percent more slow-wave sleep than men.
How well do wearables stage sleep?
| Study | Year, journal | Who and how many | Design | Main result |
|---|---|---|---|---|
| de Zambotti et al. | 2019, Behavioral Sleep Medicine | 41 adolescents and young adults | Oura (early generation) vs PSG, one night | 96% sensitivity for sleep, 48% specificity for wake; agreement 65% light, 51% deep, 61% REM; underestimated N3 by about 20 min, overestimated REM by about 17 min |
| Chinoy et al. | 2021, Sleep | 34 healthy adults, three nights | 4 wearables and 3 bedside devices vs PSG | Sensitivity for sleep 0.93 or higher; specificity for wake 0.18 to 0.54; stage results "inconsistent" |
| Miller et al. | 2022, Sensors | 53 adults, one night | Apple Watch, Garmin, Polar, Oura, WHOOP, Somfit vs PSG | Sleep vs wake agreement 86 to 89%; four-stage agreement 50 to 65% (kappa 0.20 to 0.52) |
| Altini and Kinnunen | 2021, Sensors | 106 people, 440 nights | Oura ring algorithm development vs PSG | 4-stage accuracy 57% with movement only, 79% adding heart rate, HRV, temperature and circadian features (authors employed by or advising Oura) |
| Svensson et al. | 2024, Sleep Medicine | 96 Japanese adults aged 20 to 70, up to 3 nights each | Oura Gen3 with sleep staging algorithm 2.0 vs home PSG | No significant difference for total, light or deep sleep; REM underestimated by 4 to 6 min; per-stage accuracy 75.5% (light) to 90.6% (REM); study funded by Oura |
Two patterns run through these studies. First, devices are good at detecting sleep (sensitivity above 90 percent) and poor at detecting short periods of wake (specificity often under 50 percent), so they tend to overestimate total sleep on restless nights. Second, four-stage agreement in independent studies sits around 50 to 65 percent. Newer algorithms tested by the manufacturer's own researchers report better numbers, and the newest Oura data are encouraging, but they have not yet been replicated independently at the same scale.
What that means for your morning chart
- Total sleep and timing are the most trustworthy outputs. Use them.
- Stage minutes on a single night can be out by 20 to 50 minutes, depending on device. In the 2022 study, individual night errors for deep sleep had standard deviations of 34 to 64 minutes.
- Trends in stages over weeks, from the same device, are more meaningful than any one night.
- Stage data cannot diagnose a sleep disorder. Sleep apnoea and narcolepsy need a sleep study.
The detail pages on deep sleep and REM sleep cover what each stage does and what moves it. The sleep and health guide covers how much sleep adults need.
Logging it
Peptide AI reads sleep stages from Oura, WHOOP, Garmin, Apple Health and Health Connect through wearable integrations and puts them beside the dose log, meals and training. It treats stage minutes as a trend, which is what the validation studies support.
Questions people ask
What are the stages of sleep?
Sleep labs score wake, N1, N2, N3 and REM from brain waves. Wearables usually report light (N1 plus N2), deep (N3) and REM.
How much deep and REM sleep is normal?
In a 2021 sleep lab study of healthy adults averaging 28 years, PSG recorded about 425 minutes of sleep, of which about 81 minutes was deep (19 percent) and 107 minutes REM (25 percent). Deep and REM percentages tend to fall with age.
Are sleep trackers accurate for sleep stages?
Moderately. In a 2022 study of six devices against PSG, agreement for four-stage classification was 50 to 65 percent, compared with 86 to 89 percent for simple sleep versus wake.
Why does my tracker say I slept more than I think?
Wearables detect sleep well but miss short awakenings. In a 2021 study of seven devices, specificity for wake was only 0.18 to 0.54.
Can a sleep tracker diagnose sleep apnoea?
No. Stage and oxygen data from consumer devices are not diagnostic. Sleep apnoea needs a sleep study.
Sources
- Performance of seven consumer sleep-tracking devices compared with polysomnography (Chinoy et al.), Sleep, 2021
- A validation of six wearable devices for estimating sleep, heart rate and heart rate variability in healthy adults (Miller et al.), Sensors, 2022
- The sleep of the ring: comparison of the OURA sleep tracker against polysomnography (de Zambotti et al.), Behavioral Sleep Medicine, 2019
- The promise of sleep: a multi-sensor approach for accurate sleep stage detection using the Oura ring (Altini and Kinnunen), Sensors, 2021
- Validity and reliability of the Oura Ring Generation 3 with Oura sleep staging algorithm 2.0 compared with multi-night ambulatory polysomnography (Svensson et al.), Sleep Medicine, 2024
- Meta-analysis of quantitative sleep parameters from childhood to old age in healthy individuals (Ohayon et al.), Sleep, 2004
- Normal polysomnography parameters in healthy adults: a systematic review and meta-analysis (Boulos et al.), Lancet Respiratory Medicine, 2019
- The effects of age, sex, ethnicity, and sleep-disordered breathing on sleep architecture (Redline et al.), Archives of Internal Medicine, 2004
