Quick answer
Wearable calorie counts are rough estimates, not measurements. Peer-reviewed testing against laboratory equipment has repeatedly found energy-expenditure errors of 20% to more than 90%, even from devices that track heart rate quite well. Treat the calorie number as a trend indicator for comparing your own workouts over time, not as an exact figure to plug into a strict calorie budget.
By Priya Nandakumar
Who this matters most for
This explainer is for anyone who checks the “calories burned” number on a smartwatch or fitness band after a workout and wonders whether it is trustworthy. That includes people trying to lose weight who use the figure to balance intake against output, people training for endurance events who track workout load over weeks, and casual exercisers who simply like seeing a number at the end of a session. It is written for a general audience and does not assume any background in exercise science.
It is not a buying guide for a specific watch or app, and it does not cover step counting or sleep tracking, which have their own separate accuracy profiles. The focus here is narrow and specific: how wearables estimate calorie burn, and how far off those estimates tend to be according to published research.
How wearables estimate calorie burn in the first place
No consumer wearable actually measures the calories your body burns. Measuring energy expenditure directly requires capturing the oxygen a person consumes and the carbon dioxide they produce, a process called indirect calorimetry, or tracking isotope elimination through blood, urine, or saliva samples over one to two weeks using the doubly labeled water method. Both approaches require lab equipment, trained staff, or specialized isotopes, none of which fit inside a wristband. So instead of measuring, a wearable predicts.
That prediction typically combines three or four inputs. The first is heart rate, usually captured through photoplethysmography (PPG), a technique where LEDs shine light into the skin and a sensor reads how much light bounces back as blood volume changes with each pulse. The second is motion data from an accelerometer and sometimes a gyroscope, which detects step cadence, stride pattern, and overall movement intensity. The third is a set of user-entered stats, typically age, sex, height, and weight, which feed into standard metabolic formulas such as resting metabolic rate estimates. Some devices add a fourth layer: machine-learning models trained on population-level data that try to classify what activity you’re doing (walking, cycling, elliptical, strength training) and apply an activity-specific calorie formula.
These inputs get combined through a proprietary algorithm that the manufacturer does not publish in full detail. The device essentially says: given this heart rate pattern, this movement signature, and these body statistics, similar people in our training data burned approximately this many calories per minute. It is a statistical estimate built from population averages, applied to an individual body that may or may not resemble that population.
The numbers, in brief
0 of 7
wrist devices in a Stanford cohort study kept energy-expenditure error under 20%, even though most tracked heart rate well
<10%
of energy-expenditure readings across 158 published studies met an acceptable accuracy threshold, per a 2020 JMIR systematic review
71%
of heart-rate readings from the best-performing brand landed within 3% of a chest-strap reference, the same review found
92.6%
was the highest calorie-estimate error recorded for a single wrist-worn device in the Stanford cohort study
Why calorie estimates carry more error than heart rate does
Heart rate and calorie burn sound like they should be equally reliable since one measurement (pulse) feeds into the other calculation (energy expenditure). In practice, they behave very differently, and understanding why helps explain the size of the gap.
Heart rate is a direct physiological signal. A PPG sensor detects blood volume changes with each heartbeat, and that signal, while it can be disrupted by motion artifact, skin tone, wrist tightness, or tattoos, is measuring something that is actually happening at the wrist in real time. Calorie burn is not a signal the body produces at the wrist at all. It is an inference stacked on top of an inference: the device estimates heart rate, then uses that estimated heart rate (plus motion and body stats) to predict a metabolic outcome that depends on muscle mass, fitness level, movement efficiency, terrain, temperature, and individual metabolic variability, none of which the sensor can see.
Two people with an identical heart rate during the same workout can burn meaningfully different amounts of energy. A trained cyclist and a beginner pedaling at the same heart rate are not equally efficient; the beginner’s body is working harder per heartbeat in some contexts and less hard in others, depending on fitness adaptations. Muscle mass changes the math too: someone with more lean tissue generally burns more at rest and during activity than someone of the same weight with less muscle, yet most wearables only ask for height, weight, age, and sex. None of that captures body composition, and the underlying prediction formulas were built from population averages that will always be off for individuals who sit at the edges of that population.
Activity type adds another layer of difficulty. Algorithms trained mostly on walking and running data tend to struggle when the movement pattern changes, which is a large part of why strength training and interval work produce larger errors than steady walking or running, a pattern confirmed directly in the activity-type research covered below.
What the validation studies actually found
Several independent research groups have tested wearables against laboratory-grade references, indirect calorimetry, metabolic carts, and chest-strap heart rate monitors, in controlled settings. The results are consistent across studies conducted in different years, with different devices, and by different teams.
Stanford University, 2017. A team led by Anna Shcherbina tested seven wrist-worn devices (Apple Watch, Basis Peak, Fitbit Surge, Microsoft Band, Mio Alpha 2, PulseOn, and Samsung Gear S2) on 60 participants performing walking, running, and cycling on treadmills and stationary bikes, comparing the devices against a validated reference for heart rate and against indirect calorimetry for energy expenditure. Heart-rate accuracy was fairly good for most devices, with the lowest error (1.8%) recorded during cycling and the highest (5.5%) during walking; the Apple Watch had the best overall heart-rate accuracy at a 2.0% median error. Calorie estimates told a different story entirely. No device achieved an energy-expenditure error below 20%, and errors ranged from 27.4% for the best performer (Fitbit Surge) up to 92.6% for the least accurate device tested (PulseOn). The researchers concluded that manufacturers’ proprietary formulas likely rely on limited inputs that fail to capture individual metabolic variability across different activities.
JMIR mHealth and uHealth systematic review, 2020. Daniel Fuller and colleagues pooled results from 158 published studies covering nine wearable brands. Step counting came out as the most reliable metric, with 45.2% of controlled tests landing within 3% of a reference count. Heart rate placed second, with 56.5% of tests within 3% error overall and the best brand (Apple) reaching 71% accuracy at that threshold. Energy expenditure was the clear laggard: fewer than 10% of readings met the same 3% accuracy bar, and every brand showed a consistent directional bias, some devices overestimated (Apple by 58% of readings, Polar by 69%) and others underestimated (Garmin 69% of readings, Withings 74%), meaning the errors did not even cancel out in a predictable direction across brands.
JMIR mHealth and uHealth, 2018. Ravi Reddy and colleagues put a Fitbit Charge 2 and a Garmin vívosmart HR+ through treadmill tests, stationary cycling, resistance circuits, high-intensity interval training, and daily activities, comparing them against a Polar chest strap for heart rate and Cosmed portable indirect calorimetry for energy expenditure. This study is especially useful because it breaks results out by activity type rather than reporting one blended average, which is the focus of the next section.
Healthcare Informatics Research, 2015. Fatema El-Amrawy and Mohamed Nounou reviewed 17 devices and reported a somewhat more favorable picture for step counting (79.8% to 99.1% accuracy) and heart rate (92.8% to 99.9% accuracy), concluding that the trackers were “reasonable” for estimating average activity level and average energy expenditure for general wellness purposes, while explicitly framing devices as tools that support behavior change rather than precision instruments. This earlier, more optimistic assessment is worth including because it shows the range of scientific opinion, even as it still stops short of calling the devices accurate in an absolute sense.
How accuracy changes by activity type
The clearest pattern across the research is that accuracy is not one fixed number for a given device; it shifts substantially depending on what you’re doing. Steady, rhythmic, weight-bearing cardio (walking and running on a flat surface) is the easiest condition for these algorithms because the motion pattern is repetitive and well represented in the training data manufacturers use. Cycling is harder, since the wrist barely moves relative to the amount of work the legs are doing, which strips the accelerometer of a useful signal. Resistance training and interval work are the hardest of all, because movement is stop-and-start, heart rate lags behind actual effort, and the wrist motion pattern (lifting a dumbbell, gripping a bar) does not resemble anything the walking-and-running-trained model expects.
Energy-expenditure error by activity type (Fitbit Charge 2 vs. Garmin vívosmart HR+, against indirect calorimetry)
Daily living activities (most reliable range)
Fitbit −8.8% | Garmin −10.6%
Treadmill, maximal effort test
Fitbit −17.0% | Garmin −11.4%
Free-weight resistance training
Fitbit −12.9% | Garmin +21.0%
Stationary cycling (max effort and HIIT) (least reliable range)
Fitbit −39.1% to −41.9% | Garmin −4.5% to −11.2%
Source: Reddy et al., JMIR mHealth and uHealth, 2018. Bar length reflects the magnitude of error (percent difference from indirect calorimetry); negative values mean the device underestimated calories burned, positive means it overestimated.
Notice two things in that data. First, the same activity can produce very different errors on two different devices (resistance training was under-counted by the Fitbit but over-counted by the Garmin), which means brand-switching does not reliably fix the problem, it just changes its direction. Second, cycling was the worst category for both devices, despite cycling being one of the more “controlled” cardio activities in a gym setting, which lines up with the low wrist-motion signal problem described above.
Table summarizes patterns reported in Reddy et al. (2018), Shcherbina et al. (2017), and Fuller et al. (2020); see References for full citations.
Practical implications: what to actually do with the number
None of this means the calorie figure on your watch is useless. It means the number is better suited to relative comparisons than absolute accounting. A wearable applies roughly the same estimation method to every session you log on it, so if last Tuesday’s run shows a higher calorie estimate than today’s run at a similar distance and pace, that relative difference is more trustworthy than either number taken alone. The device is consistent with itself even when it is inconsistent with reality.
Where this becomes a real problem is when the calorie estimate gets treated as a precise budget item, for example, “I burned 500 calories at the gym, so I can eat 500 extra calories tonight.” Given that published error rates commonly run from 20% to over 40%, and sometimes far higher depending on the activity and device, that 500 could easily represent anywhere from roughly 300 to 700 actual calories, or an even wider range during cycling or interval sessions. Building a tight daily calorie budget on top of a number with that much uncertainty tends to produce frustration that has nothing to do with effort or adherence; it is a measurement problem, not a willpower problem.
A more useful approach is to use wearable data as one input among several, alongside how workouts feel (perceived exertion), whether strength or endurance metrics are trending up over weeks, body measurements or how clothing fits, and, if weight change is the goal, average body weight tracked over rolling weekly windows rather than single days. The calorie estimate can still flag big swings, a much harder session than usual, a rest day, but it should not be the single number a whole nutrition plan rests on.
Checklist: using calorie estimates sensibly
- Treat the number as a trend, not a receipt; compare your own sessions over time rather than banking on the absolute figure.
- Expect larger errors during cycling, weightlifting, and interval training than during walking or steady running.
- Enter accurate height, weight, age, and sex in the app; these inputs anchor the underlying formula.
- Do not “spend” workout calories on food one-for-one; build in a buffer given typical 20% to 40%-plus error margins.
- Use average body weight over 1 to 2 week windows, not daily wearable totals, as the primary check on progress.
- If a goal depends on precise calorie numbers (competitive weight class, managing a medical condition), talk to a registered dietitian rather than relying on device output alone.
- Re-check assumptions if you switch devices or brands; the direction and size of the error can shift noticeably.
A worked example
Jordan is trying to lose a modest amount of weight and wants to run roughly a 400-calorie weekly deficit through exercise, on top of eating consistently. Jordan’s watch says a 45-minute strength training session burned 410 calories, and a 40-minute stationary bike class burned 480 calories. Based on the device totals, Jordan assumes the week’s workouts alone created almost a full day’s worth of extra deficit.
Applying what the research above suggests: resistance-training estimates can run anywhere from noticeably under to noticeably over the true figure depending on the device (a range of roughly minus 13% to plus 21% was reported in one controlled study), and stationary cycling estimates are frequently understated by 40% or more on some devices, or overstated on others. So the strength session’s real energy cost could plausibly sit anywhere from about 320 to 500 calories, and the cycling session’s real cost is even less certain, it could be notably higher or lower than the 480 shown, depending on which brand of device Jordan is wearing.
Rather than stacking these two uncertain numbers into a precise weekly deficit target, a steadier approach for Jordan is to keep the workouts consistent week to week, track body weight as a rolling average, and treat the calorie totals as a rough signal that this was an active week, not as the deciding input for how much extra to eat. Over a month, the trend in body weight and how workouts feel will say more than any single session’s calorie estimate.
Common mistakes
What exercise physiology researchers say
The consistent theme across the studies referenced in this piece is a gap between two different jobs a wearable is asked to do. Detecting a pulse at the wrist is a well-understood engineering problem with a fairly direct signal to work from, and the Stanford team’s own data reflects that: most devices kept heart-rate error under about 5% to 6% across activities. Predicting metabolic energy cost is a different kind of problem altogether, one that depends on physiology the sensor cannot directly observe, and the same research team found zero devices meeting a 20% error threshold for that specific measurement, a finding the Fuller et al. 2020 review echoed at a much larger scale across 158 separate studies.
El-Amrawy and Nounou’s 2015 review lands in a more moderate place, framing trackers as reasonable tools for gauging average activity level and supporting behavior change, without claiming precision. Read together, the body of research points to the same practical conclusion: these devices are genuinely good at nudging people toward moving more and noticing patterns in their own activity, and genuinely limited at telling anyone exactly how many calories a specific session burned.
Safety and scope notes
This article is general educational information about measurement accuracy and is not medical or nutrition advice, and it is not a substitute for individualized guidance. If you have specific caloric or nutritional needs, including a diagnosed medical condition such as diabetes, a heart condition, or a metabolic disorder, or a history of disordered eating, please consult a registered dietitian or physician before using calorie estimates from any device to guide eating decisions. Wearable calorie data should never be used to justify restrictive eating patterns, and anyone experiencing distress around food intake or exercise should reach out to a qualified healthcare provider.
Key takeaways
- Wearables estimate calorie burn using heart rate, motion data, and user-entered body stats run through a proprietary formula; they do not measure it directly.
- Heart-rate tracking is reasonably accurate on most modern devices, often within a few percent of a chest strap.
- Energy-expenditure (calorie) estimates carry much larger errors, commonly 20% to over 40%, and sometimes far higher, according to controlled laboratory studies.
- Accuracy varies sharply by activity: walking and running are the most reliable; cycling, resistance training, and interval work are the least reliable.
- Different brands err in different directions, so switching devices does not guarantee more accurate numbers, only differently wrong ones.
- Use calorie estimates as a relative, session-to-session trend indicator, not as a precise input for a strict calorie budget.
Frequently asked questions
Are Apple Watch, Fitbit, and Garmin calorie counts equally accurate?
No. Published research has found different brands err in different directions and by different amounts depending on the activity. In one 2020 review, Apple and Polar devices tended to overestimate energy expenditure while Garmin and Withings tended to underestimate it, and in a 2018 activity-specific study, Fitbit and Garmin diverged sharply on the same resistance-training sessions. No single brand came out as consistently precise across activity types.
Why is heart-rate tracking more accurate than calorie tracking on the same device?
Heart rate is a direct signal the sensor can pick up at the wrist through light-based pulse detection. Calorie burn is not directly measurable at the wrist; it is a downstream prediction built from heart rate, motion, and basic body stats, and that extra layer of estimation is where most of the error gets introduced.
Which activities give the least accurate calorie estimates?
Cycling, resistance training, and high-intensity interval training consistently show the largest errors in controlled studies, sometimes exceeding 40%. Steady walking and running tend to produce the smallest errors, though even those are commonly off by 10% to 20%.
Can I fix the accuracy by entering more detailed personal information?
Entering accurate height, weight, age, and sex helps the underlying formula start from a better baseline, and it is worth keeping updated. It does not fully close the gap, though, because most consumer devices still cannot account for body composition, fitness level, or movement efficiency, all of which affect true energy expenditure.
What is the gold-standard method researchers use to check these devices?
Indirect calorimetry, which measures oxygen consumption and carbon dioxide production during exercise, and doubly labeled water, which tracks isotope elimination over one to two weeks in free-living conditions, are the two reference methods researchers use. Both require lab equipment or specialized isotope analysis, which is why they are used to validate wearables rather than being consumer options themselves.
Should I stop using the calorie feature altogether?
Not necessarily. The feature is still useful for spotting relative trends, whether one workout was harder than another, or whether an active week looked different from a sedentary one. The research simply suggests treating the absolute number with skepticism rather than building a strict calorie budget on top of it.
Do more expensive wearables give more accurate calorie estimates?
Not consistently. The Stanford study tested devices across a range of price points and found that better heart-rate accuracy did not translate into better calorie accuracy for any of them. Price and sensor quality mostly affect signal capture, not the metabolic modeling problem, which is the harder underlying issue.
References
- Shcherbina, A., Mattsson, C. M., Waggott, D., et al. (2017). “Accuracy in Wrist-Worn, Sensor-Based Measurements of Heart Rate and Energy Expenditure in a Diverse Cohort.” Journal of Personalized Medicine, 7(2), 3. mdpi.com/2075-4426/7/2/3
- Fuller, D., Colwell, E., Low, J., et al. (2020). “Reliability and Validity of Commercially Available Wearable Devices for Measuring Steps, Energy Expenditure, and Heart Rate: Systematic Review.” JMIR mHealth and uHealth, 8(9), e18694. mhealth.jmir.org/2020/9/e18694
- Reddy, R. K., Pooni, R., Zaharieva, D. P., et al. (2018). “Accuracy of Wrist-Worn Activity Monitors During Common Daily Physical Activities and Types of Structured Exercise: Evaluation Study.” JMIR mHealth and uHealth, 6(12), e10338. mhealth.jmir.org/2018/12/e10338
- El-Amrawy, F., & Nounou, M. I. (2015). “Are Currently Available Wearable Devices for Activity Tracking and Heart Rate Monitoring Accurate, Precise, and Medically Beneficial?” Healthcare Informatics Research, 21(4), 315-320. doi.org/10.4258/hir.2015.21.4.315
- “Doubly Labeled Water.” Overview of the isotope-based reference method for measuring total energy expenditure in free-living conditions. en.wikipedia.org/wiki/Doubly_labeled_water
