Sleep-Stage Estimates: How Accurate Are They?
How closely consumer wearables match sleep-lab scoring of light, deep and REM sleep, with device-specific study results and what the agreement numbers mean.
Content last updated . This page has not been medically reviewed.
Quick answer
Consumer devices are better at telling sleep from wake than at sorting sleep into light, deep and REM stages, and agreement with a sleep lab varies by device model, software version and group tested. Labels used: established, promising but limited, mixed, insufficient evidence and Proper evidence not found (we looked and found nothing usable).
What sleep stages are and how a lab scores them
Sleep has two basic types: REM (rapid eye movement) and non-REM, and non-REM has three stages. Stage 3, the deepest, occurs in longer periods in the first half of the night, and REM sleep first occurs about 90 minutes after falling asleep. [1] A sleep lab (polysomnography) records brain waves, eye and limb movements, breathing, oxygen levels and heart rate. [1] Trained scorers then label each 30-second epoch following American Academy of Sleep Medicine criteria. [2]A wrist or ring device has no brain-wave sensor. It estimates stages from movement and heart signals using a company algorithm. Recent Fitbit models that use heart rate variability and movement performed better than early models that used movement alone. [3] So a "deep sleep" figure on a device is an estimate of a stage that a lab would define by brain activity.
How to read the accuracy numbers
- Sensitivity: how often the device finds a stage the lab found.
- Precision: how often the device's call of a stage was right.
- Cohen's kappa: agreement beyond chance, from 0 (none) to 1 (perfect). The 2025 six-device study below described its 0.21 to 0.53 range as fair to moderate agreement. [4]
- Bland-Altman limits of agreement: the range in which most nightly differences fall. Wide limits mean a single night can be far off.
What validation studies found
These results belong to the exact models and groups named. Do not carry them to other models or newer software. The table is not a ranking.| Study | Devices | People | Result |
|---|---|---|---|
| Lab study, 2024 [5] | Oura Ring Gen3, Fitbit Sense 2, Apple Watch Series 8 | 35 healthy adults aged 20 to 50, one night | Stage sensitivity ranged 50% to 86%; Oura not different from polysomnography on stage minutes; Fitbit underestimated deep sleep by 15 minutes; Apple underestimated deep sleep by 43 minutes |
| Lab study, 2025 [4] | Fitbit Charge 5, Fitbit Sense, Withings ScanWatch, Garmin Vivosmart 4, Whoop 4.0, Apple Watch Series 8 | 62 adults, mean age 46, one night | Kappa 0.21 to 0.53 (fair to moderate); specificity 29% to 52% |
| Oura Gen3 study, 2024 [2] | Oura Ring Gen3 with its 2.0 staging algorithm | 96 healthy adults aged 20 to 70, multiple nights | Stage accuracy from 75.5% (light) to 90.6% (REM) |
| Multicenter study, 2023 [6] | 11 wearables, bedside devices and phone apps | 75 people in Korea | Macro F1 score (an overall stage-agreement score) ranged from 0.26 to 0.69; the performance varied by BMI and sleep apnea severity |
| Home study, 2026 [7] | Apple Watch Series 7, Fitbit Charge 5, Polar Vantage M2 | 54 healthy adults aged 18 to 45 | Agreement generally weak with wide limits; Apple Watch overestimated REM by about 30 minutes and underestimated light sleep by about 74 minutes |
| Lab study, 2021 [8] | Four wearables, three bedside devices | 34 healthy young adults | Stage comparisons mixed; devices worse on disrupted nights |
Notice two things. Results for the same brand differ between studies, and between models and algorithm versions, so one good result does not cover a whole brand. And because people, nights, scoring and methods differ between studies, numbers from one study cannot be compared head to head with another. [6]
Who is left out
Most of these studies enrolled healthy adults. The Sleep Research Society review of wearable sleep technology says performance may drop with factors such as skin color or obesity, and that results may not transfer to people with certain conditions. [9] The 2023 multicenter study found stage agreement differed by BMI and by sleep apnea severity. [6] Label for stage accuracy overall: mixed.Accuracy is not benefit
Even a perfect stage estimate would not show that tracking improves health. The American Academy of Sleep Medicine says consumer sleep technology cannot be used to diagnose or treat sleep disorders. [10] One 2026 study concluded such devices may suit longitudinal self-monitoring rather than clinical-grade assessment of sleep architecture. [7] For what stage numbers can mean, see deep sleep and REM scores and wearables vs sleep studies. To judge a new study yourself, see how to read a wearable validation study.Frequently asked questions
Can my watch tell me how much deep sleep I got?
It can give an estimate. Studies found it can be off by many minutes compared with a sleep lab, and the direction of the error varies by device. [5]
Which tracker is most accurate?
We do not rank devices. Results come from different people, nights and methods, so they cannot be compared directly. [6]
Why do two trackers show different stages?
Each uses its own algorithm, and agreement with polysomnography ranged from weak to substantial across devices in the studies above. [7][4]
Are stage estimates good enough for health decisions?
No. Use them as rough trends at most, and see a clinician about sleep problems. [10]
More details: evidence, limits and cautions
The details behind this page: cautions, limits of the evidence and where sources disagree. Open any note to read it.
Safety note
If you snore loudly, stop breathing at night or feel very sleepy by day, see a clinician; a device stage chart is not a test for a sleep disorder. See sleep apnea.
References
- National Institute of Neurological Disorders and Stroke (NINDS). Brain Basics: Understanding Sleep. Read 2026-10-22.
- Validity and reliability of the Oura Ring Generation 3 (Gen3) with Oura sleep staging algorithm 2.0 (OSSA 2.0) when compared to multi-night ambulatory polysomnography. Sleep Medicine 2024. Read 2026-10-22.
- Accuracy of Wristband Fitbit Models in Assessing Sleep: Systematic Review and Meta-Analysis. J Med Internet Res 2019. Read 2026-10-22.
- A performance validation of six commercial wrist-worn wearable sleep-tracking devices for sleep stage scoring compared to polysomnography. Sleep Advances 2025. Read 2026-10-22.
- Accuracy of Three Commercial Wearable Devices for Sleep Tracking in Healthy Adults. Sensors 2024. Read 2026-10-22.
- Accuracy of 11 Wearable, Nearable, and Airable Consumer Sleep Trackers: Prospective Multicenter Validation Study. JMIR mHealth uHealth 2023. Read 2026-10-22.
- Performance of three consumer sleep-tracking devices compared with Actigraphy and Polysomnography. Sleep Medicine 2026. Read 2026-10-22.
- Performance of seven consumer sleep-tracking devices compared with polysomnography. Sleep 2021. Read 2026-10-22.
- State of the science and recommendations for using wearable technology in sleep and circadian research. Sleep 2024. Read 2026-10-22.
- Consumer Sleep Technology: An American Academy of Sleep Medicine Position Statement. J Clin Sleep Med 2018. Read 2026-10-22.
Education, not medical advice. This page is for general information. It does not diagnose or treat any condition and is not a substitute for care from a qualified health professional. Natural does not automatically mean safe. Ask a clinician or pharmacist before changing your care, especially if you take medication, are pregnant or breastfeeding, or have a health condition.