How Accurate Are Wearable Calorie Counts? Apple Watch, Garmin, Fitbit, and Oura Tested

Your watch’s calorie number is the least trustworthy figure it shows you. Every systematic review of the field has landed on the same verdict: consumer wrist devices do not estimate energy expenditure accurately. Stanford’s benchmark evaluation found no device among seven tested with an error under 20%, and the largest review of the literature put the error above 30% for every major brand.¹²³

The same watch tracks your heart rate with a median error under 5%.¹⁴

That gap is not a manufacturing defect. It’s the difference between a quantity your watch measures and a quantity it guesses — and once you see which is which, the whole dashboard reads differently.

Here’s what the labs found when they put wrist devices next to indirect calorimetry, and what the number on your wrist is actually good for. It isn’t nothing. It’s just not what the ring closing implies.

How accurate are wearable calorie counts, really?

Not accurate enough to plan a diet around. The published error rates, measured against metabolic carts and doubly labelled water:

Study	Devices tested	Energy expenditure finding
Stanford, 2017 ¹ (n=60)	Apple Watch, Fitbit Surge, Samsung Gear S2, + 4 more	No device under 20% error. Apple Watch best, Samsung Gear S2 worst
Systematic review, 2020 ² (158 studies)	9 brands	”For energy expenditure, no brand was accurate”
Systematic review, 2022 ³ (65 studies)	All major wrist brands	MAPE >30% for every brand
Field test, 2022 ⁵ (n=20)	Apple Watch S6, Garmin FENIX 6, Huawei GT 2e	9.9%-32.0% error; Apple and Garmin both overestimated
Lab test, 2026 ⁶ (n=62)	Apple, Galaxy, Fitbit, Garmin	EE underestimated in endurance; near-random in lifting

The 2022 McMaster University review is the blunt one. Federico Germini and colleagues screened 65 validation articles and concluded that while the Apple Watch tracked heart rate to under 10% error and Fitbit tracked steps to under 25%, “none of the tested devices proved to be accurate in measuring energy expenditure.”³

Read that as a class verdict, not a brand problem. Switching watches does not fix it.

If you’re weighing which device to buy, the honest framing is that they differ on the metrics that are reliable — a comparison we work through in detail in our Apple Watch vs Oura vs WHOOP vs Garmin breakdown. Calorie accuracy shouldn’t be your tiebreaker, because no one wins it.

Why is heart rate accurate but calories aren’t?

Because your watch actually measures one of them.

Heart rate comes from photoplethysmography: green LEDs pulse light into your wrist, a photodiode reads how much comes back, and blood volume changes produce a signal your watch counts. It’s an imperfect physical measurement of a real, physical event happening two millimetres under the sensor.

Calories are not measured. They’re inferred — a model that takes heart rate, motion from the accelerometer, and whatever profile data you typed in (age, weight, sex, height), and produces an estimate of oxidative metabolism happening in tissue the watch cannot see.

Think of it like judging a car’s fuel consumption from the engine noise. The noise is real and correlated. But the conversion from sound to litres per hundred kilometres depends on the engine, the load, the gearing, and the fuel — none of which the microphone knows.

The gold standard, indirect calorimetry, works by measuring the oxygen you consume and the carbon dioxide you produce. That’s the actual chemistry of energy release. A wrist sensor has access to neither gas.

So the error compounds: heart rate error (small) feeds a population-average model (large error) applied to an individual whose real metabolic economy may sit well outside the average. And crucially, the manufacturers don’t publish those algorithms, which means neither you nor a researcher can audit why a given number came out the way it did.

This is the distinction that shapes how SensAI reads your data: signals the hardware genuinely measures get weight, and modelled outputs get treated as soft context rather than ground truth.

Which device is worst? Apple Watch, Garmin, Fitbit, and Oura

Error direction is not consistent between brands — which is why two watches on the same wrist will disagree with each other and with the lab.

Apple Watch and Garmin tend to overestimate during cardio. A Shanghai Jiao Tong University team ran 20 adults through outdoor walking at 6 km/h and running at 10 km/h against a COSMED K5 portable metabolic cart. The Apple Watch Series 6 overshot by 19.8% walking and 24.4% running. The Garmin FENIX 6 was worse walking — 32.0% error, with an intraclass correlation of just 0.216, effectively no agreement — and 21.8% running.⁵

Fitbit tends to underestimate. A meta-analysis of 52 Fitbit validation studies pooled 29 energy-expenditure comparisons and found an average underestimate of 2.77 kcal per minute, with limits of agreement running from −12.75 to +7.41 kcal/min.⁷ Over a 45-minute session, the plausible range spans hundreds of calories.

Oura validates well in the lab and still drifts in the wild. The first validation of Oura Ring energy expenditure, from the University of Gothenburg, found a strong laboratory correlation against indirect calorimetry (r = 0.93) — though even there the ring tended to underestimate, with the gap widening as intensity rose. Across a 14-day free-living study with 32 participants, its total energy expenditure differed from reference monitors by 362 to 494 kcal.⁸

The lesson from those three paragraphs: a 400-calorie discrepancy is normal, expected, and brand-dependent in direction. If you eat back what your watch says you burned, that error lands directly in your energy balance.

It’s also why SensAI never asks you to reconcile two devices’ calorie figures. Whichever wearable you’ve connected, the useful comparison is against your own previous sessions on that same hardware.

Why calorie error gets worse when you lift

Cardio is the easy case for these algorithms. Steady heart rate, rhythmic motion, predictable relationship between the two. Resistance training breaks all three assumptions.

The most recent evidence is the clearest. A 2026 study from Yonsei University’s Exercise Physiology Laboratory put 62 healthy adult men through standardized endurance and resistance protocols wearing four smartwatches simultaneously — Apple, Galaxy, Fitbit, and Garmin — with ECG and indirect calorimetry as references.⁶

Heart rate held up beautifully. Correlations with ECG ran 0.64 to 0.97, reliability was excellent (ICC > 0.94), and limits of agreement sat around ±10 bpm. During resistance exercise specifically, only the Apple Watch showed no significant difference from ECG.

Energy expenditure fell apart. During endurance work it was consistently underestimated with wide limits of agreement. During resistance training, correlations with indirect calorimetry collapsed to r = 0.10-0.34 and reliability dropped below ICC 0.45.⁶

A correlation of 0.10 is, for practical purposes, unrelated to the truth.

The mechanism is intuitive once you picture it. A heavy set of five is enormously metabolically demanding but involves almost no wrist movement and a heart-rate response that lags the effort and peaks after you rack the bar. The accelerometer sees stillness. The optical sensor is fighting muscle-tension artefact. The model, trained mostly on ambulatory activity, has nothing useful to work with.

So if your training is mostly barbell work, your watch’s calorie figure is closer to a random number than an estimate. That’s worth knowing before you use it to decide whether your program is producing results.

For strength sessions, SensAI judges the work by load, reps, and how your recovery responds over the following days — the things that actually moved — rather than by an estimate the sensor was never positioned to make.

Whose readings are worst? The accuracy gap nobody discloses

The Stanford study that set the benchmark also found the error isn’t evenly distributed.

Anna Shcherbina and colleagues, working under Euan Ashley — professor of medicine and genetics at Stanford and one of the founders of the university’s precision health programme — tested seven devices on 60 volunteers of deliberately varied age, height, weight, skin tone, and fitness across sitting, walking, running, and cycling.

Six of seven devices measured heart rate to a median error below 5% during cycling. None reached a sub-20% error on energy expenditure.¹

And the errors were systematically larger for men, for people with higher body mass index, for people with darker skin tone, and during walking.¹

That last set of findings matters more than the headline. A metric that’s wrong is one problem. A metric that’s differently wrong depending on your body composition and skin tone is another, because it means the population you belong to determines how misled you are — and nothing in the interface tells you which group you’re in.

Newer work continues to probe optical sensor performance across Fitzpatrick skin types, and the direction of the field is toward acknowledging the gap rather than closing it.⁹

The deeper problem: “calories out” isn’t a fixed number anyway

Suppose the sensor were perfect. You’d still be misreading the result, because the underlying model of how exercise affects daily energy expenditure is wrong.

The intuitive model is additive: burn 400 calories on the bike, and your daily total goes up by 400. Herman Pontzer, professor of evolutionary anthropology at Duke University, has spent a decade demonstrating that humans don’t work this way.

In a 2026 Current Biology analysis, Pontzer and Eric Trexler compared additive and constrained models against experimental and free-living data. Their finding: in human aerobic exercise interventions, total daily energy expenditure rose by only about 30% of the change the additive model predicted.¹⁰ The body quietly claws back the rest — trimming basal and sleeping metabolic rate, and reducing spontaneous movement.

Two nuances worth carrying:

Compensation appeared reduced with resistance training — one more argument for lifting if body composition is the goal.¹⁰
Compensation was amplified when aerobic exercise was paired with diet restriction — precisely the combination most people attempt.¹⁰

So the watch tells you 400. The physiology delivers something closer to 120 in net daily terms, and does so on a delay you can’t observe. Stack a 25% sensor error on top of a 70% compensation effect and “eat back your exercise calories” stops being an approximation and becomes a reliable way to stall.

If you’re setting intake targets, work from bodyweight-anchored math and adjust on observed trend — the approach we lay out in how many calories you should actually eat — not from a number your wrist invented.

What should you actually do with the calorie number?

Don’t delete the metric. Demote it.

Use it as a relative workload marker, not an absolute quantity. Your watch’s error is reasonably consistent for you, at a given activity type. So 500 today versus 300 on Tuesday is real information about relative effort, even though neither figure is metabolically true. Compare the number only to your own history, on the same device, for the same kind of session.

Never compare across devices or across people. A Garmin 500 and an Apple Watch 500 are different claims about the world, and the 2022 field data shows they can differ by more than 12 percentage points of error on the same activity.⁵

Never eat it back at face value. If you use it at all for intake, discount it heavily — the compensation literature suggests something in the region of a third of face value is closer to the net daily truth.¹⁰

Lean on the metrics that survived validation. Heart rate is solid. Steps are solid. Sleep duration and resting heart rate trends are usable. These are the inputs worth building decisions on, and they’re the ones our wearable data accuracy guide treats as load-bearing.

Watch the trend, not the reading. A single day’s figure carries mostly noise. A fortnight’s direction carries signal.

This is exactly the reasoning SensAI applies to connected health data: it reads heart rate, recovery, and sleep trends from Apple Watch, Garmin, Oura, and WHOOP through HealthKit, and weights them by how well each signal actually validates — rather than treating every number on the dashboard as equally true. A modelled calorie estimate and a measured heart rate do not deserve the same confidence, and a coach that can’t tell the difference will give you confident advice built on the weaker one.

Frequently asked questions

Are Apple Watch calories accurate?

No, though the Apple Watch is among the better performers. Stanford’s evaluation found it had the lowest overall energy-expenditure error of seven devices tested — but still above the 20% threshold, with no device clearing it.¹ Field testing put its error at 19.8% for walking and 24.4% for running, in the direction of overestimation.⁵

Does the Apple Watch overestimate or underestimate calories?

During outdoor walking and running, it overestimates — significantly so against a metabolic cart.⁵ During resistance training, smartwatch estimates as a class tend to underestimate, and correlate so weakly with true expenditure (r = 0.10-0.34) that direction is close to meaningless.⁶

How many calories does a fitness tracker get wrong?

Expect 20-30% error as the realistic floor, and more in the field. For a session your watch reports as 500 calories, a plausible true range is roughly 350-650. Fitbit’s pooled limits of agreement span −12.75 to +7.41 kcal per minute, which over an hour is a several-hundred-calorie window.⁷

Is Garmin or Apple Watch more accurate for calories?

In the head-to-head outdoor test, the Apple Watch Series 6 was more accurate than the Garmin FENIX 6 for walking (19.8% vs 32.0% error), while the Garmin was slightly better for running (21.8% vs 24.4%).⁵ Neither is accurate in absolute terms, and the ranking flips by activity — which is itself the argument against trusting either figure.

Is the Oura Ring accurate for calories?

Better than most in a laboratory setting (r = 0.93 against indirect calorimetry), but it underestimates more as intensity rises, and across 14 days of free-living use its total energy expenditure differed from reference monitors by 362-494 kcal.⁸

Should I eat back my exercise calories?

Not at face value. Two independent errors stack: the sensor overstates or understates the session, and your body compensates for roughly 70% of the added expenditure over the following days.¹⁰ If you eat back the full number, you’re likely erasing the deficit you created.

Which wearable metrics can I actually trust?

Heart rate (typically under 5-10% error in validated devices), step count (Fitbit Charge models under 25% error across 20 studies), and sleep and resting-heart-rate trends.¹³⁴ Energy expenditure is the outlier, and it’s the metric most fitness apps lean on hardest.

Will calorie accuracy improve?

Slowly. Meta-analysis shows that combining heart rate or heat-flux sensing with accelerometry reduces error meaningfully, which is the direction the hardware is moving.¹¹ But the fundamental problem — inferring tissue-level metabolism from the outside of a wrist — is not solved by a firmware update.

The bottom line

Your watch measures heart rate. It guesses calories. No systematic review of the major brands has yet concluded that the guess is accurate — across nine years and hundreds of published validations.¹²³

Treat the calorie figure as a rough relative effort marker on a single device, compared only against your own history. Treat heart rate, steps, and multi-week recovery trends as the real data. And when you’re deciding what to change in your training, anchor on what your body is actually doing — strength progressing, sessions tolerated, recovery holding — rather than on a modelled number that a metabolic cart would contradict by 400 calories.

The wearable’s job is to capture signal. Deciding what that signal means for your next session is a different job entirely, and it’s the one worth getting right.

References

Shcherbina A, Mattsson CM, Waggott D, Salisbury H, Christle JW, Hastie T, Wheeler MT, Ashley EA. “Accuracy in Wrist-Worn, Sensor-Based Measurements of Heart Rate and Energy Expenditure in a Diverse Cohort.” Journal of Personalized Medicine, 2017; 7(2): 3. https://pubmed.ncbi.nlm.nih.gov/28538708/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸
Fuller D, Colwell E, Low J, Orychock K, Tobin MA, Simango B, Buote R, Van Heerden D, Luan H, Cullen K, Slade L, Taylor NGA. “Reliability and Validity of Commercially Available Wearable Devices for Measuring Steps, Energy Expenditure, and Heart Rate: Systematic Review.” JMIR mHealth and uHealth, 2020; 8(9): e18694. https://pubmed.ncbi.nlm.nih.gov/32897239/ ↩ ↩² ↩³
Germini F, Noronha N, Borg Debono V, Abraham Philip B, Pete D, Navarro T, Keepanasseril A, Parpia S, de Wit K, Iorio A. “Accuracy and Acceptability of Wrist-Wearable Activity-Tracking Devices: Systematic Review of the Literature.” Journal of Medical Internet Research, 2022; 24(1): e30791. https://pubmed.ncbi.nlm.nih.gov/35060915/ ↩ ↩² ↩³ ↩⁴ ↩⁵
Nelson BW, Allen NB. “Accuracy of Consumer Wearable Heart Rate Measurement During an Ecologically Valid 24-Hour Period: Intraindividual Validation Study.” JMIR mHealth and uHealth, 2019; 7(3): e10828. https://pubmed.ncbi.nlm.nih.gov/30855232/ ↩ ↩²
Le S, Wang X, Zhang T, Lei SM, Cheng S, Yao W, Schumann M. “Validity of three smartwatches in estimating energy expenditure during outdoor walking and running.” Frontiers in Physiology, 2022; 13: 995575. https://pubmed.ncbi.nlm.nih.gov/36225296/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶
Lee TH, Jun DU, Bae JY, Roh HT, Cho SY. “Comparative Validity of Smartwatch-Derived Heart Rate and Energy Expenditure During Endurance and Resistance Exercise.” Sensors, 2026; 26(8): 2526. https://pubmed.ncbi.nlm.nih.gov/42076635/ ↩ ↩² ↩³ ↩⁴
Chevance G, Golaszewski NM, Tipton E, Hekler EB, Buman M, Welk GJ, Patrick K, Godino JG. “Accuracy and Precision of Energy Expenditure, Heart Rate, and Steps Measured by Combined-Sensing Fitbits Against Reference Measures: Systematic Review and Meta-analysis.” JMIR mHealth and uHealth, 2022; 10(4): e35626. https://pubmed.ncbi.nlm.nih.gov/35416777/ ↩ ↩²
Kristiansson E, Fridolfsson J, Arvidsson D, Holmäng A, Börjesson M, Andersson-Hall U. “Validation of Oura ring energy expenditure and steps in laboratory and free-living.” BMC Medical Research Methodology, 2023; 23(1): 50. https://pubmed.ncbi.nlm.nih.gov/36829120/ ↩ ↩²
Kostrna J, Oparina E, Palacios C, Rodriguez AJ, Pei J, Ajmal A, Ramella-Roman JC. “PPG-Based Heart Rate Accuracy in Hispanic Adults with Fitzpatrick III-V Skin Tones: An Evaluation of Body Composition and Skin-Tone Effects.” Sensors, 2026; 26(10): 2922. https://pubmed.ncbi.nlm.nih.gov/42197732/ ↩
Pontzer H, Trexler ET. “The evidence for constrained total energy expenditure in humans and other animals.” Current Biology, 2026; 36(4): 1013-1025.e4. https://pubmed.ncbi.nlm.nih.gov/41653928/ ↩ ↩² ↩³ ↩⁴ ↩⁵
O’Driscoll R, Turicchi J, Beaulieu K, Scott S, Matu J, Deighton K, Finlayson G, Stubbs J. “How well do activity monitors estimate energy expenditure? A systematic review and meta-analysis of the validity of current technologies.” British Journal of Sports Medicine, 2020; 54(6): 332-340. https://pubmed.ncbi.nlm.nih.gov/30194221/ ↩