How accurate are wearable calorie (active energy) burn estimates?

This sits alongside two other accuracy questions we've covered separately: how reliable Apple Watch's VO2max estimate is (a different metric, similar theme) and how many steps per day actually matter (a much more reliable volume metric than calories). This guide is specifically about the number your wearable calls "calories burned" or "active energy," and how much weight it can actually bear.

How big is the error, really?

The most-cited reference point is a 2017 Stanford study that tested seven wrist devices — including Apple Watch, Fitbit Surge, and Samsung Gear S2 — against a metabolic cart in a lab setting. Six of the seven devices measured heart rate within about 5% of the reference value. None of them measured energy expenditure well: errors ranged from roughly 27% (the best performer) to 93% (the worst), and the error grew substantially for cycling and mixed-modality workouts compared with steady walking or running.

More recent outdoor testing, using a portable indirect-calorimetry unit instead of a treadmill-based metabolic cart, tells a similar story with somewhat smaller numbers, which is what you'd expect as algorithms have improved over the past decade:

Two patterns hold up across nearly every study in this space, regardless of brand or year: heart rate is measured well, and calorie burn is not — and the error consistently gets worse for activities where heart rate and motion don't track effort in a simple, steady way.

Why calorie burn is so much harder to estimate than heart rate

Your wearable's optical or electrical sensor is picking up an actual physical signal when it measures heart rate — a pulse of blood or an electrical impulse it can detect directly, verify against, and average over a few seconds. There's no equivalent sensor for "calories." Instead, the device runs your heart rate, accelerometer data, and the age/sex/height/weight you entered through a proprietary formula that was calibrated on a reference population and then applied to you.

That formula is doing two jobs at once, and only one of them is genuinely hard:

• Resting energy expenditure (the calories you'd burn lying still, sometimes called basal metabolic rate) is fairly predictable from body size, age, and sex using well-established equations, and wearables are reasonably consistent here.
• Active energy — the calories attributable specifically to movement and exercise — depends on exercise economy, muscle mass, fitness level, and exactly how efficiently your particular body converts oxygen into motion at a given heart rate. None of that is visible to a wrist sensor, so the algorithm is extrapolating from population averages, and any one person can sit meaningfully off that average in either direction.

This is also why "total calories burned today" and "active calories from this workout" are genuinely different numbers with different reliability — the first is dominated by the fairly predictable resting component, while the second is almost entirely the harder-to-estimate active component, which is where most of the error concentrates.

Body composition changes the error, too

A 2026 study that tested four popular smartwatches (Apple Watch Series 8, Samsung Galaxy Watch 5, Garmin Forerunner 955, and a Fitbit model) against indirect calorimetry found average errors in the 15-25% range — but the error grew larger as a participant's body fat percentage increased, regardless of brand. The likely explanation is that the underlying sensor signal and the population data used to build the algorithm don't represent the full range of real body compositions equally well. The same study looked for a skin-tone effect and didn't find a clear one, though the sample of participants with the darkest skin tone was too small to rule it out with confidence — worth flagging honestly rather than treating either finding as settled.

Apple Watch's overestimation was consistently the smallest of the four devices tested in that study; Samsung and Garmin's were the largest. That's broadly consistent with the 2017 Stanford findings and the 2022 outdoor study, which is one reason Apple Watch estimates are used as a reference point through this guide — not because they're accurate in an absolute sense, but because they're the most-validated and typically least-biased of the mainstream options.

Which activities give you the least reliable number

Not all workouts are equally hard to estimate. The pattern that shows up across studies:

• Best case: steady-state cardio at a stable heart rate — treadmill walking, jogging, or an elliptical at consistent effort. Heart rate and motion both track effort in a simple, linear way, which is the exact scenario the underlying algorithms were built and validated on.
• Worse: cycling. The cardiovascular effort is real, but sitting reduces the accelerometer signal wrist devices lean on, and pedaling doesn't create the same wrist-swing pattern as walking or running — so accuracy drops even though your heart rate reading itself may still be fine.
• Worst: strength training and other stop-start activity. Heart rate spikes during a heavy set and drops during the rest between sets — a pattern that looks, second to second, nothing like the steady elevated heart rate of a cardio session, even when total energy cost is similar. One controlled study found wearables overestimated resistance-training energy expenditure by an average of roughly 40%, and the mechanism (short bursts of high effort, heart rate lag, minimal wrist accelerometer signal from a squat or bench press) is well understood even where exact numbers vary by study.

What the number is actually good for

None of this means the "calories burned" figure on your wrist is useless — it means it's useful for a narrower job than most people ask it to do. It's a reasonable relative signal: this week's total was meaningfully higher or lower than last week's, or this run felt harder and the number reflects that. It's a much weaker absolute number, and two of them stacked together — wearable calories burned minus a food-tracking app's calories eaten — compound each source's 15-40% error into a deficit estimate that can be wrong by a much larger margin than either number alone suggests. If you're trying to size how much protein you actually need for your training, body weight and training type are a sturdier starting point than a precise calorie-deficit calculation built on wearable numbers.

The most reliable inputs from wearable data remain the ones with a direct physiological signal behind them — heart rate itself, resting heart rate, HRV, sleep, and step count — which is part of why Vita's recovery score and Body Age model weight those over derived estimates like calorie burn. If you're trying to figure out whether a number you're seeing reflects a real change in your training or just normal device noise, Vita's AI coach can walk through your own trend alongside the more reliable signals rather than treating any single calorie figure as ground truth.