How Heart Rate Estimates Calories Burned

Your pulse is a stand-in for oxygen, and oxygen is the thing being measured#

A watch does not measure calories from your heart rate. A calorie is a unit of heat, and nothing on a wrist captures heat or gas exchange. What the device captures is a pulse. That pulse is pushed through an equation that predicts how much oxygen you are consuming, and the oxygen figure is multiplied by a conversion factor — around five kilocalories per litre, the textbook value, which itself shifts by a few percent depending on whether you are burning mostly fat or mostly carbohydrate. One measurement, two models stacked on top of it.

The chain is real physiology, not marketing. Within one person, across a decent band of aerobic effort, heart rate and oxygen uptake climb together closely enough to be useful, and researchers were estimating expenditure this way decades before wearables existed. The catch is who the line was fitted for. In the cleanest free-living test of the method, 14 adults were measured simultaneously by doubly labeled water over 15 days and by heart-rate monitoring: the two methods produced almost identical group averages, 12.99 MJ/day against 12.89 — while individual heart-rate estimates ran from 22.2 percent below the isotope figure to 52.1 percent above it, with only nine of the fourteen landing within ±10 percent¹. The authors' own verdict: the method "provides a close estimation of the TEE of population groups." Whether the resulting calories are worth spending at all is the pillar's question. This article is about how the number gets built, and the four places the build fails.

The chain, link by link#

Step	What actually happens	What it assumes about you
Wrist sensor → heart rate	Green light reads the blood-volume pulse through skin	The optical signal is clean during motion
Heart rate → oxygen uptake	A regression fitted to a research sample	Your pulse-to-oxygen line matches theirs
Oxygen uptake → calories	Multiply by a caloric equivalent of oxygen	Your fuel mix is near the assumed mix

Only the first row is measurement. The second row is where nearly all the error enters, and it is worth seeing what a good version of that regression actually needs. When 115 regularly exercising adults aged 18 to 45 were tested at 35, 62 and 80 percent of VO₂max, the model that best predicted expenditure from heart rate required sex, heart rate, weight, VO₂max and age together, and reached a correlation of 0.913 with measured expenditure — 83.3 percent of the variance. Strip out VO₂max, which almost no consumer device knows, and the same model drops to r = 0.857 and 73.4 percent².

That gap is the price of not knowing your fitness. Two people of the same age, sex and weight with different aerobic capacities have genuinely different pulse-to-oxygen lines: the fitter one does the same work at a lower heart rate, so an equation blind to fitness will systematically read their effort wrong. Your watch estimates VO₂max, at best, from the same heart-rate data it is trying to interpret.

Right for the room, wrong for the person in it#

The most useful way to read this literature is to notice that the validations agree with each other and still do not say what people think they say.

Setting	Reference method	How the group mean did	How individuals did
Whole-body calorimeter, 21 h³	Indirect calorimetry	−1.2% (SD 6.2), not significant	Range −11.4% to +10.6%
Free-living, 15 days¹	Doubly labeled water	12.99 vs 12.89 MJ/day	Range −22.2% to +52.1%
Steady-state exercise⁴	Indirect calorimetry	Overestimated by roughly 6–18%	Depended on activity and model

Read the last two columns against each other. In a sealed metabolic chamber with individually calibrated curves, heart rate is genuinely good — a mean error of about one percent and a worst case near eleven. Walk the same method out into ordinary life and the group average stays excellent while the individual spread more than triples. Nothing about the method got worse; the conditions did. Real days contain heat, stress, caffeine, illness, standing in queues and sitting in traffic, and a pulse responds to all of them.

Hilloskorpi's team, testing 87 adults, concluded that at minimum sex and body weight must be in the equation before heart rate can predict activity expenditure at all — and their predicted values still overran the measured ones by roughly 6 to 18 percent depending on the activity and the model used⁴. Each of these papers is a validation. None of them is a promise about you.

Your heart rate climbs while your metabolism sits still#

The assumption the whole chain rests on is that a rising pulse means rising work. Over a long session, that quietly stops being true.

Nine cyclists rode at 60 percent of VO₂max in 35 °C heat. Between minute 15 and minute 45 — same bike, same workload — heart rate rose from 151 to 169 beats per minute, a 12 percent climb, while stroke volume fell 16 percent, from 120 to 101 mL per beat. Submaximal oxygen uptake increased only slightly. What did change sharply was the relative intensity: because VO₂max itself fell 19 percent over the same half hour, the riders went from working at 63 percent of maximum to 78 percent without touching the pedals differently⁵.

This is cardiovascular drift, and it is the plainest illustration of why pulse is a proxy rather than a meter. The heart compensates for a falling stroke volume by beating more often. A model watching only the beats reads that compensation as extra work and charges you calories for it. The longer and hotter the session, the more of your burn readout is drift rather than metabolism — which is one reason a long summer ride and the same ride in October produce different totals for identical effort. (What a bike ride actually costs has the one instrument that sidesteps this entirely, by measuring mechanical work at the pedals.)

At low intensity, the pulse carries no information at all#

The other end of the range fails differently, and the failure is old enough to have a name. Below a threshold, heart rate stops tracking oxygen uptake usefully: sitting, standing, a stressful email and a slow walk can all produce a similar pulse and very different metabolic costs. Researchers handle this with a flex heart rate — a personal cut-off above which the calibration curve is applied and below which expenditure is simply assumed to equal resting metabolism.

In the calorimeter study above, that threshold sat at 86 ± 10 beats per minute in men and 96 ± 6 in women³. Livingstone's free-living group averaged 97 ± 8¹. Consider what that means for a normal day: for the fifteen or so waking hours you spend below that line, the heart-rate method is not estimating your expenditure from your heart rate. It is assigning you a resting value and moving on. The pulse data is discarded because it is known to be uninformative there.

Lifting weights breaks the chain outright#

Resistance training is the case where the relationship does not merely loosen — it stops existing. Sixty-two healthy adult men wore four commercial smartwatches through standardized endurance and resistance protocols against indirect calorimetry. For heart rate, the devices were strong: correlations of 0.64 to 0.97 and reliability above an ICC of 0.94, with agreement limits around ±10 beats per minute. For energy expenditure during the resistance protocol, the same devices produced correlations of 0.10 to 0.34 and ICCs below 0.45⁶.

One device, one session, two numbers: the pulse is nearly perfect and the calories are close to noise. The physiological reason is that a heavy set drives heart rate up through pressure and effort against a braced trunk rather than through sustained oxygen demand, and then most of the set's real metabolic cost is paid afterwards, during recovery between sets. The equation sees a high pulse in the moment and a low one in the rest interval, and neither matches the actual oxygen debt. What a lifting session genuinely costs is worked through in calories burned strength training.

What genuinely improves the estimate#

Two things measurably help, and neither turns the number into a budget line.

The first is adding a second signal. When 46 adults wore a combined heart-rate and movement sensor for 14 days against doubly labeled water, acceleration alone under-read activity expenditure by 12.1 kJ/kg/day (r = 0.52) and heart rate alone came in near-unbiased but imprecise (RMSE 34, r = 0.58). Combining them through branched equation modelling cut the root-mean-square error to 20 and lifted the correlation to 0.67, a statistically significant improvement over heart rate alone⁷. Movement and pulse fail in different situations — acceleration is blind to a hill, pulse is blind to the difference between a hill and an argument — so together they cover more of the day than either does alone.

The second is calibrating the curve to you rather than to a sample, which improves precision in the same study. That is what a proper lab test buys, and it is why a chest strap plus a measured VO₂max sits meaningfully closer to the truth than a wrist reading a stock equation. It is also why the accuracy record of consumer devices looks the way it does across brands and activities, which how accurate fitness-tracker calories are documents in full.

What none of it delivers is a number precise enough to eat against. The most useful posture is the one the original researchers took: heart rate is an excellent instrument for comparing groups, a decent one for comparing you to yourself on similar sessions, and a poor one for stating what a particular Tuesday cost — the same reason calorie figures are better read as ranges than as totals.

FAQ#

Why does my heart rate drift upward when my pace hasn't changed?#

Because your stroke volume is falling and the heart is compensating with frequency. In cyclists holding a constant workload in the heat, heart rate rose 12 percent between minute 15 and minute 45 while stroke volume dropped 16 percent and oxygen uptake barely moved. Your calorie readout treats that climb as extra work, so a long or hot session inflates the total without any additional metabolic cost.

Can I make my watch's calorie estimate more accurate?#

Partly. Entering your correct height, weight, age and sex matters, because the underlying equations are built on exactly those terms — and adding a fitness measure matters more: the best heart-rate model explained 83.3 percent of the variance in expenditure with VO₂max included and 73.4 percent without it. A chest strap plus a device that also senses movement helps again, since combining the two signals cut prediction error by roughly 40 percent against doubly labeled water. None of that makes the figure precise enough to eat back.

Why are calorie estimates so unreliable for weight training?#

Because heart rate and oxygen demand come apart during resistance work. Across four smartwatches tested on 62 men, heart-rate accuracy held up during lifting (ICC above 0.94) while energy-expenditure estimates fell to correlations of 0.10 to 0.34 with laboratory measurement. A heavy set raises pulse through pressure and bracing rather than sustained aerobic demand, and much of the real cost is paid in the rest intervals the model reads as recovery.

Sources#

This article was researched and drafted with AI assistance and reviewed for accuracy by the BurnWeek team. It is general information, not medical advice. How we research and correct our articles →