Is the Apple Watch accurate?

Why this question has a better answer for Apple than for anyone else

Most of this market is unmeasured. A 2024 umbrella review of 24 systematic reviews, covering 249 primary validation studies and 430,465 participants, found that roughly 34 of 310 identified consumer wearables have ever been validated at all. The usual honest answer to “is this device accurate” is that nobody has checked.

Apple is the exception. Four peer-reviewed results we could read in full involve an Apple Watch: a sleep-lab comparison, two laboratory VO2 max validations and a distance study. No other brand on this site has that much published about it. That is worth saying before the findings, because the findings are mostly unflattering and there is an easy misreading here — the Apple Watch does not look worse than its rivals because it is worse. It looks worse because somebody measured it and did not measure them.

Sleep: the best coarse call of the three devices tested, and the worst fine one

Robbins et al. (2024) at Brigham and Women's Hospital put 35 healthy adults through a single night of inpatient polysomnography wearing an Apple Watch Series 8, a Fitbit Sense 2 and an Oura Ring Gen3 simultaneously. For the Apple Watch:

• Sleep versus wake: 93.0% agreement(κ = 0.60) — the highest of the three devices
• Four-stage classification: 75.0% agreement(κ = 0.60), just behind the Oura Ring's 76.3% and ahead of the Fitbit's 70.9%
• Deep sleep sensitivity: 50.5%— the weakest result of the three. It identified about half the deep sleep that was actually happening
• Nightly summaries: light sleep overestimated by 45 minutes, deep sleep underestimated by 43 minutes, wake underestimated by 7 minutes

Read those four lines together and a shape appears that holds across everything else on this page. Asked a coarse question — is this person asleep — the watch is excellent. Asked a fine one — which stage of sleep is this — it moves roughly three quarters of an hour of deep sleep into the light-sleep column. The nightly total you see when you wake up is broadly trustworthy. The colored bands underneath it are the part that moved.

One methodological detail the authors report and we will not bury: six of the Apple Watches failed during the study. That is a real limitation of the dataset, and it belongs beside the results rather than in a footnote. The full cross-device table is on our sleep tracking accuracy page.

VO2 max: measured twice in a laboratory, and short both times

Apple Watch VO2 max is the only consumer VO2 max estimate with published validation against laboratory gas analysis that we could read. Two studies, two protocols, the same direction of error.

Lambe et al. (2025, PLOS ONE) tested 28 participants across Series 5-9, SE 2 and Ultra 2 on watchOS 10 or later against indirect calorimetry. The watch underestimated by 6.07 mL/kg/minon average, a mean absolute percentage error of 13.31%, with limits of agreement from −6.11 to +18.26. The authors' own conclusion was that the estimates are not sufficiently accurate to inform clinical decision-making.

Caserman et al. (2024, JMIR Biomedical Engineering) tested a Series 7 against a portable gas analyzer in 19 people: laboratory 45.88 against watch 41.37 mL/kg/min, a bias of −4.51, MAPE 15.79%, and an intraclass correlation of 0.47 that the authors describe as poor reliability.

The most useful finding in either paper is what happened when the second study split its sample by fitness. In the fittest subgroup (n = 5) the watch read 42.70 against a measured 54.70 — MAPE 21.47%. In the least fit subgroup (n = 3) it overestimated, 38.93 against 35.13, MAPE 10.71%.

That is a systematic error, not noise. The estimate pulls toward the middle of the distribution: if you are fit it will tell you that you are less fit than you are, and the fitter you get the further it drifts. For the buyer this site is written for — someone already training seriously and shopping for better data — that is precisely the wrong direction for the error to run. Our VO2 max accuracy page carries every row, and our explainer on what the number is covers why an estimate from heart rate and pace can only ever be an inference.

Note the subgroup sizes: five people and three people. Those are small enough that the exact percentages should be held loosely. The direction is what the two studies agree on.

Distance and GPS: one measurement, from 2020, on hardware two generations old

The eight-watch distance validation (Gilgen-Ammann et al., 2020) measured an Apple Watch Series 4 against a trundle wheel accurate to 1 cm across urban, forest and 400 m track conditions. The Apple Watch came in within the study's overall 3.2-6.1% error band; only Polar's watches finished under 5% across every condition.

That result is six years old and describes a single-frequency receiver. Apple's own published Ultra 3 specification states L1 and L5 precision dual-frequency GPS across GPS, GLONASS, Galileo, QZSS and BeiDou — different hardware doing a different job, and nobody has published a measurement of it. So the honest position on current Apple GPS is: the engineering is among the best specified in the category, and the validation record is silent. The same is true of every current dual-band watch, which is the point of our GPS accuracy comparison.

Heart rate: the question most people are actually asking

Here is the gap. None of the studies we have read in full measures an Apple Watch's optical heart rate against an ECG reference during exercise. We are not going to quote a figure we have not read, so this page quotes none.

What we can say comes from two places. The umbrella review puts consumer wrist heart rate at approximately ±3% accuracy overall, with a mean bias of −3.39 bpm across the pooled literature — a category-level figure, not an Apple one, and one that flatters steady-state conditions because that is where most validation happens.

And the 2025 anatomical placement study (n = 28, Polar H10 ECG reference) tested devices at the wrist, forearm and upper arm through rest, cycling, burpees and a treadmill ramp. No Apple device was in it — say that out loud — but its finding is about the measurement site rather than the brand: the wrist was the weakest position tested, and the errors appeared during rapid intensity change rather than at steady effort. An optical sensor on the back of a wrist is reading blood flow through a joint that bends, slides and tightens, and no brand has engineered its way out of that.

So: if your Apple Watch tracks your easy run and disagrees wildly during intervals, that is the expected behavior of the hardware class, not a defective unit. Our wrist versus chest strap page has the physics, and our Garmin equivalent of this page shows the same pattern on the other dominant brand — including a wrist watch that recorded roughly 33 bpm of bias during burpees.

Steps and calories, briefly

Neither has an Apple-specific validation in our source set, and the category figures from the umbrella review are worth knowing anyway: step count mean absolute percentage error from −9% to 12%, and energy expenditure with a mean bias of −3 kcal per minute and errors ranging from −21.27% to 14.76%.

Calories are the least reliable number any wearable shows you, by a wide margin, on every brand. An error of −3 kcal per minute compounds to roughly 180 kcal an hour. Treat the move ring as a consistency tool — same watch, same activity, is today more than yesterday — and never as a measurement you can eat against.

What the fix actually fixes

For activity heart rate there is a fix, it is cheap, and it is the same one Apple's own ecosystem supports: pair an external sensor over Bluetooth and the watch records from that instead of from your wrist. The full ranking of straps that pair with an Apple Watch covers the field; the three below are the picks that matter most.

Be clear about the boundary, though, because it is where most buying advice on this topic quietly oversells. A chest strap fixes exercise heart rate. It does not fix your sleep staging, your VO2 max estimate, your calorie figures or your overnight HRV, because nobody wears a chest strap to bed and the watch computes those from its own sensor. If the number that frustrates you is a sleep stage or a fitness estimate, no accessory on this site will improve it.

How to read any Apple Watch number

One rule carries almost all of the published evidence above: trust the coarse call, distrust the fine one.

• Total sleep last night— trustworthy. 93.0% agreement on asleep-versus-awake is a good number by any standard.
• Your deep sleep minutes— not trustworthy as a quantity. The watch found half of it in the one study that checked, and moved most of the rest into light sleep.
• Your VO2 max, as a number— expect it to read low if you are fit, by 4.5 to 6 mL/kg/min on the measured averages and more at the top end.
• Your VO2 max, as a trend— more useful. A systematic error that runs in a consistent direction still moves when your fitness moves.
• Steady-state heart rate on an easy run— broadly fine, on the category evidence.
• Heart rate during intervals, lifting or anything with rapid changes — this is where wrist optical sensing fails, and where a strap earns its money.
• Calories— the weakest number on any wearable. Use it for relative comparison only.

And one framing worth keeping: the Apple Watch has more published evidence behind it than anything else we rank, and that is a reason to trust the brand's transparency position rather than a reason to distrust the product. A device with known errors is a better instrument than a device with unknown ones. Our evidence hierarchy is built on exactly that principle, and our Apple Watch Ultra 3 review applies it to the current flagship.