Why Your Smartwatch's Calorie Count Is Off By Up to 93%

Educational journalism, not medical advice. Every claim here is checked against its cited sources by editor Tim Bunce — a health writer, not a physician. It isn’t specific to your situation: for health decisions, talk to your own clinician. How we work →

What the validation studies found

A landmark 2017 Stanford study (Shcherbina et al., published in the Journal of Personalized Medicine) tested seven wrist-worn devices of that era — Apple Watch, Basis Peak, Fitbit Surge, Microsoft Band, Mio Alpha 2, PulseOn, and Samsung Gear S2 — against indirect calorimetry, the laboratory gold standard for measuring energy expenditure. Subjects performed treadmill walking, treadmill running, and stationary cycling while metabolic gas analysis ran in parallel.

Heart rate measurement was generally accurate: most devices fell within 5–10% of chest-strap values, with errors concentrated during high-intensity intervals where wrist optical sensors lose contact.

Calorie expenditure was a different story. Median errors ranged from 27% for the most accurate device (Fitbit Surge) to 93% for the least accurate (PulseOn) — Apple Watch, despite leading on heart rate, still missed calorie estimates by about 40%. Across all seven devices, none produced energy-expenditure estimates within the 20% error threshold the researchers flagged as the bar for reasonable accuracy.

Why the algorithm fails

The math chain a wearable uses to estimate calories:

1. Read the optical heart rate signal from the wrist.
2. Apply a smoothing algorithm to clean noise.
3. Calculate VO₂ estimation from heart rate using a regression model.
4. Convert VO₂ to energy expenditure using a respiratory quotient assumption.
5. Apply a body-weight scaling factor based on user-entered demographics.
6. Output a calorie number.

Each step compounds error. The biggest contributors:

• The VO₂-from-heart-rate regression assumes a population-average relationship. Individual variation is substantial. The same heart rate can correspond to dramatically different actual oxygen uptake across people.
• The respiratory quotient assumption uses a fixed value. In reality RQ varies with what fuel the body is burning (carbohydrate vs. fat) and shifts with intensity.
• The body-weight scaling doesn’t account for body composition. A 70 kg person at 12% body fat and a 70 kg person at 28% body fat have very different actual energy expenditure for the same activity.
• Resistance training is particularly bad because heart rate elevation underestimates energy expenditure during anaerobic work.

What wearables actually measure well

The same studies show wearables are accurate for several metrics that matter more for training than calorie counts:

• Heart rate at steady-state exertion: within 3–5% of chest strap for most devices, most of the time.
• Step count: typically within 5% of actual.
• Sleep duration: reasonably accurate to within ~15 minutes. Sleep stage detection is less reliable.
• Resting heart rate trends over time: highly reliable. Changes in resting HR are a strong recovery and overtraining signal.
• HRV (heart rate variability) trends: useful when interpreted as relative change, not absolute value.

What metrics to actually trust

The practical hierarchy for trusting wearable data:

1. Trend, not absolute number. Your resting HR trending up over a week signals real change; your calorie count yesterday doesn’t.
2. HR and HRV during sleep. The night-time data is the cleanest signal because motion artifacts are minimized.
3. Steps and active minutes. The simple metrics are the most accurate.
4. Heart rate during steady-state activities. Walking, jogging at moderate pace, cycling at moderate pace — these are the wearable’s sweet spot.
5. Calorie expenditure: don’t trust the number. Use it as a relative comparison day-to-day, not as an absolute budget.

Implications for training decisions

The practical adjustments if you’ve been using wearable calorie data to guide eating or training:

• For weight management: stop trying to match calorie intake to wearable calorie output. The output number is likely off by 20–90%. Better approach: weigh weekly and adjust calorie intake by ~200 kcal/day based on trend. Slower, more accurate.
• For training intensity: use heart rate zones, not calorie burn, to gauge effort. The HR data is reliable; the calorie translation is not.
• For recovery: use resting HR trend and HRV trend, not the wearable’s “readiness” or “strain” scores. The component metrics are more reliable than the composite scores built on top.
• For sleep: trust the total duration metric, treat the stage detection as approximate, and trust your subjective feeling more than the device’s sleep score.

Relative device rankings from the published studies

From most-accurate to least-accurate for calorie estimation in the 2017 Stanford study:

1. Fitbit Surge: most accurate of the seven devices on energy expenditure, still 27% off.
2. Apple Watch: most accurate on heart rate, but 40% off on calorie estimates.
3. Basis Peak, Microsoft Band, and Mio Alpha 2: fell in the middle of the pack, each missing energy expenditure by a wide margin.
4. Samsung Gear S2: among the least accurate on both heart rate and calorie estimation.
5. PulseOn: least accurate of the seven on calorie estimation, off by 93%.

The honest summary

Wearables are valuable for trends, sleep, HR, and step counts. They’re not accurate enough to guide calorie-based eating decisions. The mainstream advice of “eat in a deficit relative to your wearable’s calorie output” is built on data that’s off by enough to undermine the decision.

The 5–10 year horizon for this category looks better. Chest-strap HR is already accurate; the algorithm side is improving as devices add more sensors (skin temperature, blood oxygen, etc.). Future generations will likely close the gap. Current generation has not yet.

Practical takeaways

• Calorie estimates from wearables are off by 27–93% across activities. Don’t use them as absolute budgets.
• Heart rate and step count are accurate. Trust those.
• Trends, not snapshots. Resting HR going up over a week is real; yesterday’s number alone is not actionable.
• Sleep duration is reliable; sleep stages are approximate; composite “readiness” scores are noisy.
• For weight management: weekly weigh-in trend + 200 kcal adjustment beats matching calorie input to wearable output.
• For training: use HR zones, not calorie burn, to gauge effort.
• Even the most accurate device tested (Fitbit Surge) was off by 27%, and the Apple Watch — despite leading on heart rate — still missed calorie estimates by about 40%.

Related: Smartwatch vs Chest Strap for Heart Rate · HRV Readiness Metrics.

References

Additional sources reviewed for this article: Shcherbina 2017, Fuller 2020, Cvetkovic 2024.