How accurate is Apple Watch calorie tracking, and why heart rate is a different story

Short answer

Calorie estimates are the least accurate number on your watch. In a Stanford study that compared seven wrist devices against clinical gas analysis, no device estimated energy expenditure to within 20% error, and median errors across tasks ran from 27.4% to 92.6%. A systematic review of 158 publications reached the same conclusion in one line: "For energy expenditure, no brand was accurate."

Heart rate is a different story. In the same Stanford study the Apple Watch's median heart rate error was 2.0%, and the review found all brands measured heart rate to within ±3% on average in controlled settings.

If you have been quietly wondering why the calories never seem to add up — why the watch says 2,400 and the treadmill says 700 for the same hour and your weight does something unrelated to both — you were not imagining it and you are not bad at reading your own device. The number is genuinely soft. That is a property of the measurement, not a failure of yours.

Note: this article is educational and not medical advice. No consumer wearable measures calories; every device estimates them from an algorithm. Nothing here is a basis for setting a calorie deficit, and disordered eating or unexplained weight change needs a clinician rather than a better tracker.

Why a watch cannot measure calories at all

Start with what is actually happening on your wrist, because the gap makes sense once you see it.

The clinical way to measure energy expenditure is indirect calorimetry: you breathe into a mask, and an instrument measures the oxygen you consume and the carbon dioxide you produce. Burning fuel consumes oxygen in a fixed relationship, so gas exchange gives a real physical measurement of how much energy your body is spending right now.

Your watch has no access to any of that. It has a green light and a photodiode, which read the pulse of blood under your skin, plus an accelerometer that reads movement. From those two signals plus what you typed into the setup screen — height, weight, age, sex — a proprietary equation produces a calorie figure.

So the calorie number is not a measurement with some error attached. It is a prediction, and one built from inputs that carry only indirect information about metabolism. Two people the same height, weight and age, doing the same movement at the same heart rate, can genuinely burn different amounts. The watch has no way to know which of them you are.

The Stanford authors put the mechanism plainly: since height and weight are fixed and heart rate is now measured accurately, the remaining error comes either from equations that do not use heart rate at all, or from real person-to-person variation in how much energy a given activity costs.

What the validation studies found

Two studies are worth knowing about, and they agree.

The Stanford study

Researchers at Stanford tested seven wrist-worn devices — Apple Watch, Basis Peak, Fitbit Surge, Microsoft Band, Mio Alpha 2, PulseOn and Samsung Gear S2 — on 60 volunteers (29 men, 31 women, aged 38 ± 11), deliberately recruited across a range of age, height, weight, skin tone and fitness.

The reference standards were the real ones: continuous 12-lead ECG for heart rate, and continuous clinical-grade indirect calorimetry for energy expenditure. Participants sat, walked at two speeds, ran at two speeds and cycled at two intensities.

The authors set their acceptable error threshold at 5%.

Heart rate results: lowest error on the cycle ergometer, 1.8% across devices; highest on walking, 5.5%. Across all devices and activities the Apple Watch had the lowest heart rate error at 2.0% and the Samsung Gear S2 the highest at 6.8%.

Energy expenditure results: median error across tasks ranged from 27.4% for the Fitbit Surge to 92.6% for the PulseOn. Error was lowest on walking (31.8%) and running (31.0%) and highest when sitting (52.4%). The authors' summary: none of the devices estimated energy expenditure within an acceptable range in any setting.

That sitting figure deserves a second look. The activity where you burn the most total calories over a day — being alive and mostly still — is where the estimate is furthest off.

The systematic review

A separate team reviewed 158 publications covering nine device brands and 5,934 participants, sorting every comparison by whether it fell within ±3% of the reference in lab conditions.

For energy expenditure, no brand was within ±3% more than 13% of the time. Pooling everything, 9.2% of comparisons landed inside ±3%, 54.1% were more than 3% below the reference and 36.7% more than 3% above.

The direction of the miss varied by brand rather than the size of it. Garmin devices underestimated 69% of the time and Withings 74%; Apple devices overestimated 58% of the time and Polar 69%. Fitbit split roughly evenly, underestimating 48.4% and overestimating 39.5%.

For heart rate the same review found the opposite picture: Apple Watch within ±3% 71% of the time, Fitbit 51%, Garmin 49%, with the conclusion that all devices tested were within acceptable measurement error for heart rate.

There is one finding here that cuts against intuition and is worth stating: the review found no relationship between having a heart rate sensor and estimating calories better. Adding pulse to the equation did not rescue it.

What about Fitbit, Garmin, Whoop and Oura?

People usually arrive at this question about a specific device, so it is worth saying what the evidence covers and what it does not.

Fitbit is the most studied brand by a distance — 144 of the review's studies, across 12 models. It recorded the lowest median calorie error in the Stanford comparison at 27.4%, and a later meta-analysis looked specifically at combined-sensing Fitbits against reference measures and again found energy expenditure outside acceptable limits. The Stanford authors offered a plausible reason Fitbit's estimate held up relatively better: its algorithm appears to use a published resting metabolic rate equation rather than a purely proprietary one. Relatively better still means badly.

Garmin was covered by 42 studies across 13 models, and the pattern there was consistent underestimation — below the reference in 69% of comparisons.

Whoop and Oura are where the honest answer is that there is much less published validation for energy expenditure specifically. Neither appears in the two studies above. Both have validation literature for other measures — sleep staging and overnight heart rate variability, mainly — and the absence of calorie validation is not evidence that they are worse. It is an absence, and it should be reported as one rather than filled in with a guess.

The general finding holds regardless: none of this is a ranking, and the same equation that overestimates for one person can underestimate for another. The brand on your wrist is a much smaller factor than which number you are reading.

What this does and does not mean

Two honest caveats before anyone builds a conclusion on these numbers.

The devices are old. The Stanford study ran on 2017-vintage hardware, and Basis Peak, Microsoft Band and Mio Alpha 2 no longer exist. Algorithms have moved on. What has not changed is the physics: a wrist sensor still cannot see oxygen consumption, so the ceiling on how good a prediction can get is set by the inputs, not by the code.

Lab is not life. Both studies mostly measured people on treadmills and ergometers, which is the friendliest possible case — steady, known, sustained activity. Free-living days, with their carrying and stair-climbing and standing about, are harder, and the review's free-living numbers were no better.

And one thing the studies do not say, which is easy to misread. The Stanford paper found heart rate error was higher for men, for higher BMI and for darker skin tone. Those findings are about heart rate, not calories. It would be wrong to extend them to calorie estimates, and we are not going to.

We are also not going to tell you which brand is best. The published data shows different brands miss in different directions, all by a lot, on hardware mostly no longer sold. Anyone ranking them from this evidence is adding an opinion that the numbers do not contain.

What the calorie number is actually good for

Here is the useful part, because "it's inaccurate" is not advice.

A consistent error and a random error are very different problems, and the calorie figure mostly has the first kind. If your watch overestimates your Tuesday run by 30%, it will overestimate next Tuesday's run by roughly 30% too. That makes the absolute number nearly useless and the comparison between your own days still meaningful.

So:

Reasonable to do with it. Compare this week's activity to your own last month. Notice that Thursday was a much bigger day than Friday. Watch a trend move in a direction over a season.

Not reasonable to do with it. Set a calorie deficit. Eat back exercise calories. Compare your number to your friend's number — different device, different algorithm, different body, no shared scale. Conclude anything about your metabolism from a single figure.

The trap that catches most people is the third one on that list: eating back the calories the watch reports. If the estimate is 30% high and you eat to it, you have quietly turned a fitness tracker into a source of a real error in the one direction you were trying to avoid.

There is a version of this that is simply more honest: treat the calorie figure as an effort score with a misleading unit. It goes up when you do more. It is not a currency.

What you can trust more

If the calorie number is the softest thing on your wrist, what is firm?

Heart rate is the strong one, by a wide margin — around 2 to 5% error in the studies above, and within ±3% on average across brands in controlled conditions. It gets worse during movement that jostles the watch, better when you are still, and better during running than during walking, oddly, because walking involves more wrist swing relative to the signal.

Resting heart rate, measured overnight, is firmer still, because the hardest condition for an optical sensor — motion — is absent, and the measurement is averaged over hours rather than caught in a moment.

Heart rate variability inherits that accuracy when measured at rest, and carries real information about your autonomic state that a calorie estimate does not.

Steps sit in the middle: 45.2% of lab comparisons within ±3%, with a general tendency to undercount.

The pattern is consistent and worth internalising. The closer a number is to what the sensor physically detects, the more you can trust it. Heart rate is nearly direct. Heart rate variability is one step removed. Steps are a pattern-matching problem. Calories are several inferences past the last real observation, which is why they land where they do.

Things that genuinely change the estimate

A few of the levers people reach for do something, and a few do nothing. Worth separating them.

Correcting your profile does help, a bit. Height, weight, age and sex feed the equation directly, so a weight figure three years out of date propagates straight into every calorie total. This is the one adjustment with a clear mechanism.

Wearing the watch properly helps the heart rate, and therefore anything downstream. Snug, a finger's width above the wrist bone, tightened for workouts. A loose band lets ambient light reach the photodiode and lets the sensor slide, and a heart rate reading taken from a jostling sensor is worse than one taken from a still one.

Calibrating outdoor walks and runs — going out with GPS on for a few sessions — improves the device's model of your stride and pace, which improves the movement half of the input.

What does not help: recording an activity twice to "capture" more, picking a workout type that sounds harder than what you did, or switching brands in the hope of finding the accurate one. None of these gets the device closer to your oxygen consumption, which is the thing it cannot see.

And one note for anyone comparing their watch to a treadmill or a gym machine. Neither is the reference. Cardio machines estimate calories too, from speed, incline and whatever weight you typed in, and in lab comparisons they miss substantially as well. When the two disagree, that is not one accurate device and one inaccurate one. It is two estimates.

How to bring this up with your doctor

Most people never need to. But if you have taken a concern to an appointment and been shown the tracker's summary as evidence one way or the other, it helps to know which parts of your data carry weight.

What to bring. Your resting heart rate trend over weeks or months, not a single day. Overnight heart rate. Any recorded ECG strip if your device takes them. These are the parts of a consumer wearable clinicians actually find informative, and a shift in resting heart rate over time is a legitimate observation to put in front of someone.

What to leave at home. The calorie totals, the activity ring closures, and anything the app calls a score. Those are manufacturer composites, not clinical measures, and presenting them can make a real concern look less serious than it is.

How to phrase it. "My resting heart rate has been about eight beats higher than my own baseline for the past three weeks" lands very differently from "my watch says my recovery is bad." The first is an observation about you over time. The second is a brand's opinion.

If you are dismissed and your own tracked baseline has genuinely shifted and stayed shifted, it is reasonable to ask directly what would need to be true for it to be worth investigating, and to ask for that to be written in the notes. Unexplained weight change, in particular, is a clinical question and not a tracking question — no adjustment to a calorie target substitutes for finding out why.

How Welltory handles this

We are going to be straightforward about our own position here, since we are in this business too.

Welltory does not make hardware and does not estimate your calories. We read what your device already writes into Apple Health or Google Fit, and we measure heart rate variability directly — either from what your wearable recorded or through your phone's camera.

That choice follows from the same logic as the rest of this article. We focus on the measurements that sit closest to the sensor, because those are the ones worth building anything on. Welltory records your resting heart rate and HRV over time and shows you your own baseline and how far today sits from it, which is the comparison the data can actually support.

What Welltory cannot tell you: how many calories you burned, how much you should eat, or whether you are in a deficit. Nothing on your wrist can tell you those things, and a second app reading the same sensor would not change that.

See what your own heart rate data shows over time

Welltory is a general wellness product. It holds no regulatory clearance, does not diagnose, and is not a substitute for medical care.

The short version

The calorie figure is the weakest number your watch produces, by a wide margin, in every study that has checked it against clinical measurement — and the weakness is structural, not a bug that a firmware update will fix. The heart rate figure next to it is one of the strongest.

Which is a genuinely useful thing to know, because it tells you where to look. The number everybody watches is the one to trust least, and the one sitting quietly underneath it is the one carrying the information.

Data analysis by Jane Smorodnikova

Written by Tatsiana Yashyna

Discounts for blog readers: up to 36% off

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

Explore your data