How Accurate Is Your Smartwatch? What Wearable Health Sensors Get Right — and Where They Fall Short

Your smartwatch is a good heart-rate monitor, a passable sleep tracker, and a poor calorie estimator — and it does one medical screening job surprisingly well

That is the short version of what validation studies say about the numbers on your wrist. The longer version matters more, because every reading comes with fine print: how it was measured, under what conditions, and how much confidence it deserves. Researchers have now put the big brands through lab-grade testing — against chest-strap ECGs, hospital polysomnography, and metabolic carts — and the pattern is consistent. Here is the sensor-by-sensor breakdown.

Heart rate: the measurement wearables do best

Most watches read your pulse with photoplethysmography (PPG): green LEDs shine into your skin, and a sensor tracks how the reflected light changes as blood pulses through. It is an indirect measurement, which is why conditions matter so much.

In a Stanford study of 60 volunteers wearing seven devices, six of the seven measured heart rate with less than 5% error against a clinical ECG. The best performer was off by just 2.0% overall. Accuracy was strongest on a stationary bike (1.8% median error) and weakest while walking (5.5%) — arm swing is the enemy of optical sensors. The same study found that skin color and body mass index measurably affected the readings.

Consumer lab tests tell the same story with current models. Which? ran 58 devices against medical-grade chest straps: the most consistent brand averaged 1.1% error overall (under 1% at rest and while walking, 1.8% while running), while the weakest averaged 3.7%. One category stood out for the wrong reason — smart rings averaged a 20.2% error on resting heart rate, far worse than any wrist device.

The practical takeaway: a steady resting reading deserves your confidence. A sudden spike to 190 during a sprint deserves a second look — and a chest strap, if you train by heart-rate zones.

Irregular-rhythm alerts: the screening job they take seriously

Detecting atrial fibrillation (AF) is the one health feature with large-trial evidence behind it. In the Fitbit Heart Study, published in Circulation in 2022, 455,699 people were monitored by an algorithm that flags at least 30 minutes of irregular rhythm during periods of inactivity. About 1% received a notification. Among those who then wore a week-long ECG patch, the algorithm's positive predictive value for AF happening at the same time was 98.2% — consistent across age groups and sexes.

There is important nuance. AF comes and goes, so only about one in three notified people happened to have an episode during their single patch week. The study's own conclusion is careful: a notification identifies people likely to have AF, and it takes a doctor's ECG to confirm it.

Apple's Heart Study, published in the New England Journal of Medicine in 2019, enrolled more than 400,000 participants. Roughly half a percent received irregular-pulse notifications over about four months, and about 84% of the individual irregular readings matched AF on a simultaneously recording ECG patch.

So if your watch ever flags an irregular rhythm, take it seriously enough to book a checkup — but not as a diagnosis.

Blood oxygen: a wellness estimate, not a medical reading

Wrist SpO2 uses red and infrared light — the same PPG family as heart rate, with the same vulnerabilities. And there is a documented bias worth knowing about: pulse oximeters can read falsely high in people with darker skin, because melanin absorbs the light the sensor depends on.

The FDA has been moving on this. Its 2013 guidance asked pulse-oximeter makers to test on subjects with a range of skin pigmentations; a draft guidance issued in January 2025 goes further, calling for larger and more diverse study populations and the Monk Skin Tone scale to categorize participants.

One caveat: that guidance covers medical pulse oximeters, not your watch. Smartwatch SpO2 is a wellness feature, not an FDA-cleared measurement. Cold hands, motion, and poor circulation degrade it further. Use it to watch trends if you like — never to judge whether you are sick.

Sleep: good at "asleep or awake," shaky on the details

Sleep tracking is where confidence drops fastest. A 2025 study in Sleep Advances put six popular devices — including models from Fitbit, Garmin, Withings, Whoop, and Apple — through a sleep lab against polysomnography, the wired-up gold standard. The watches caught more than 90% of actual sleep, but their specificity was only 29% to 52%: lying still while awake routinely gets scored as sleep. Agreement with the lab on sleep stages was only fair to moderate. The best device managed a kappa score of 0.53; the weakest, 0.21.

A separate 2025 study in PLOS One tested 98 sleep-clinic patients and found the watch overestimated total sleep by about 29 minutes a night while underestimating time spent awake after falling asleep by 37 minutes. In other words, the watch flatters your sleep.

Trust the multi-week trend in how long you sleep. Do not rearrange your life over last night's deep-sleep minutes.

Calories burned: the number to trust least

Back to that Stanford study: not one of the seven devices estimated energy expenditure within 20% of the metabolic-cart measurement. The best missed by 27% on average; the worst by 93%. The researchers' explanation is blunt — each brand uses its own proprietary formula, and the formulas rest on assumptions about your body that may not fit you at all.

Treat the calorie figure as a rough activity index, not a budget you can eat against.

Steps: the most reliable number on your wrist

Step counting is refreshingly honest. Controlled tests put phone and wrist step counts within roughly 7% of a human observer's count, and wrist trackers under 5% error in structured conditions. A 2024 review of the literature found average errors between −9% and 12%.

It breaks in predictable situations: slow shuffling steps, pushing a stroller (one lab study measured more than 33% undercounting), or using a walker with the device on the wrist. For "did I move today," though, it is the most trustworthy metric the watch offers.

A practical trust map for your wrist data

Trust the trend: resting heart rate over weeks, nightly sleep duration over weeks, daily step counts. Patterns beat single readings every time.

Act on it, then verify: an irregular-rhythm notification. Book a checkup promptly — and let the ECG have the last word.

Enjoy with skepticism: single-night sleep-stage breakdowns, exact calorie counts, heart-rate spikes mid-sprint, and SpO2 as a verdict on your health.

Give the sensor a fair chance: wear it snug, a finger above the wrist bone; warm up cold hands before a spot check; stay still for on-demand readings; and know that motion, loose fit, and darker skin all dim the optical signal the watch depends on.

The bottom line

Watches measure the easy things well — a pulse at rest, footsteps — and estimate the hard things loosely: calories, sleep architecture. Their one genuine screening success is flagging possible atrial fibrillation, and even that needs a clinician to mean anything. Use the trends, question the digits, and keep the watch in its lane: a capable fitness companion, not a medical device.

Sources

• • Shcherbina et al., Stanford University — heart rate and energy expenditure validation of seven wearables (Journal of Personalized Medicine, 2017): medicalxpress.com
• • Lubitz et al., the Fitbit Heart Study — irregular rhythm detection in 455,699 participants (Circulation, 2022): pmc.ncbi.nlm.nih.gov
• • Fitbit Heart Study results presented at AHA Scientific Sessions 2021: healio.com
• • Apple Heart Study coverage — 84% tachogram concordance figure: gadgetsandwearables.com
• • FDA draft guidance on pulse oximeter performance across skin pigmentation (January 2025): raps.org
• • Six consumer sleep trackers vs. polysomnography (PMID 40303381, Sleep Advances, 2025): pubmed.ncbi.nlm.nih.gov
• • Wrist-worn sleep tracker validation in sleep-clinic patients (PLOS One, 2025): journals.plos.org
• • Which? lab tests of smartwatch heart-rate accuracy vs. chest straps: which.co.uk
• • Step-count and calorie accuracy roundup across studies: charitymiles.org