Wrist heart-rate sensors are getting better, but they still disagree during exercise

An optical heart-rate reading from a smartwatch or fitness band is trustworthy at rest and during easy effort, and increasingly good at hard effort — but the specific device still matters, and the error can be wide enough to mislead someone using the number to train or to gauge intensity. Three 2026 validation studies, each checking wrist devices against an electrocardiogram reference, land on the same practical point: interpret a wrist reading as device-specific, not interchangeable [s1][s2][s3].

The measurement problem is physical. Wrist sensors use photoplethysmography — shining light into the skin and reading the pulse from changes in reflected light — and motion during exercise disturbs that optical signal in a way a chest strap's electrical reading is immune to. It is the same sensor whose known limits we examined in a wrist-worn aging clock built on photoplethysmography.

Devices disagree, head to head

A study in Physical Activity and Nutrition put four consumer devices on 37 analysed participants at the same time as a Polar H10 chest strap, across low-, moderate- and high-intensity treadmill exercise [s1]. Accuracy generally improved as intensity rose, but the devices differed significantly from one another: mean absolute error varied by device across intensities (p = 0.002) [s1]. A chest-worn optical sensor (Polar Verity Sense) and one wristwatch showed significantly lower error than a Polar Pacer Pro and a Galaxy Watch 6, and the authors concluded that wearable heart-rate data should be interpreted in light of device-specific validation rather than treated as uniformly reliable [s1].

At maximal effort, a small bias but wide limits

A separate study tested the Samsung Galaxy Watch 6 on 55 healthy adults through a graded treadmill test to exhaustion, again against a Polar H10 [s2]. Agreement was not uniform across the effort range: the intraclass correlation was moderate-to-good at low and moderate intensities but dipped to 0.54 in the 70%–80% zone before recovering to 0.90 near maximal effort [s2]. Absolute error was small in the middle of the range — a median around 1–3 beats per minute, with median percentage error falling to 0.60%–0.75% at higher intensities [s2].

The caveat sits in the spread. The watch consistently read low, with an overall mean bias of −2.67 bpm, and its 95% limits of agreement ran from −16.90 to +11.57 bpm — a band roughly 28 beats wide [s2]. That average is fine for comparing groups in a study; the wide limits mean an individual reading on a given moment can be off by well over ten beats, which is enough to place someone in the wrong training zone.

The long trend is upward

A systematic review of 47 studies traced how far wrist trackers have come [s3]. Early optical monitors before 2010 sometimes reported correlations against a reference as low as 0.50 during exercise; modern devices typically exceed 0.80 during exercise, reflecting better sensors and motion-artefact filtering [s3]. Step counting has improved even more, now commonly below 5% error for steady-state walking [s3].

Energy expenditure is the laggard. Even recent devices show 10%–40% error against indirect calorimetry, because estimating calories burned without measuring oxygen and carbon dioxide is a fundamentally harder inference than counting a pulse [s3]. That is the same ceiling behind the calorie overestimates we reported in budget fitness bands, and it is why a "calories burned" figure deserves far more scepticism than a heart rate.

Why intensity, counterintuitively, can help

A striking detail in both studies is that accuracy often improved as effort rose. In the head-to-head comparison, mean absolute error and percentage error generally decreased with intensity, and concordance rose [s1]. In the maximal test, percentage error fell from 2.90% in the easiest zone to well under 1% at higher intensities [s2]. Part of the reason is arithmetic — a fixed beat-count error is a smaller percentage of a high heart rate — but a harder, steadier effort also produces a stronger, more regular pulse for the optical sensor to lock onto. The devices tend to struggle most not at peak effort but at the transitions and the awkward middle, where the 70%–80% zone in the maximal study showed the weakest agreement of all [s2].

The wrist location is its own handicap. Blood flow at the wrist is more easily disrupted by wrist flexion, grip and cold than at the fingertip or the chest, which is why the chest-worn optical sensor in the comparison study outperformed the wristwatches [s1]. A device's own validation, on its own wrist placement, is therefore the number that matters — not a rival's specification [s1].

What it means for a user

For heart rate, the reassuring finding is that the trend line and the resting number are reliable on a modern device, and accuracy tends to hold or improve at higher effort [s2][s3]. The caution is that a single reading can carry a wide error, and two brands on two wrists will not necessarily agree, so a wrist number should not be used to make a fine-grained call — an exact peak heart rate, a precise zone — that a chest strap or clinical monitor is built to make [s1][s2]. Anyone using the device for a medical reason should treat it as a prompt to check properly, not as the measurement itself.