If you are trying to judge wrist heart rate monitor accuracy across models, the first trap is treating "wrist monitor" as one sensor with one error figure. A study of one watch, on one activity, against one reference, tells you about that combination. It does not transfer to a different model, and it does not transfer to a different kind of workout.
This guide shows how to line up public evidence for the wrist device you own or are considering. It covers how to record the device and setup, match the activity and reference, check what validation exists, and keep a comparison-limits table. It is not a ranking, and it does not claim any wrist monitor is accurate or inaccurate in general.
Step 1: Record exactly which device and setup you are comparing
Write down, before you look at any numbers:
- Model and generation. Variants of the same product line, including different case sizes, are separate listings, and a different product line is a different sensor package again.
- Firmware and app version, if you can see them, because algorithm updates can change behavior after a study was run.
- Wrist, position and fit. The validation studies cited below placed devices just above the wrist bone, and one rotated devices between wrists to balance left/right effects (lab validation study of two wrist devices; transient-states study, 2025).
- Where the number comes from: the watch screen, a recorded activity file, or a phone app that merges several sources. If more than one device is involved, identify which device produced each heart-rate sample before you compare.
This process will not tell you whether a particular reading on your wrist was an artifact or true physiology. It only helps you decide how much a comparison can support.
Step 2: Match the activity and the reference
Heart rate error depends heavily on what the arm is doing, so compare like with like.
- Steady cardio versus resistance work. One lab study recruited 29 recreationally active adults aged 18–26 and compared two wrist devices with a chest strap. Both wrist devices agreed well with the strap at rest. During treadmill walking and running at 5, 8 and 10 km/h, one wrist device agreed well and the other did not, and both had poor accuracy during five resistance exercises (Physical Activity and Health). That is two specific devices, one young sample, one reference.
- Incremental treadmill running. A 2026 study had 18 physically active adults (12 men, 6 women) wear one wrist watch during an incremental treadmill test. Mean absolute percent error for heart rate was 13.0% (±10.1) against a pre-set acceptable threshold of under 5%, with proportional bias present, so the authors called heart-rate accuracy unacceptable in that test (Applied Sciences). It applies to that watch, that protocol and that sample.
- Changing intensity. A 2025 study tested wrist wearables against a 12-lead ECG in a 20-minute protocol with transitions (sitting to standing, walking speed and incline changes) and steady states (Sensors). Use it as a reminder that a watch can look different in steady segments than around a change in pace. Check its tables for which devices were included before applying any figure to yours.
The reference matters too. A chest strap and a 12-lead ECG are not the same standard, and a second wrist device is not a reference at all. If you compare two watches on your own arms, you learn how they differ from each other, not which is right.
Identical workloads in different sessions also do not guarantee identical true heart rate. Sleep, heat, caffeine and fatigue can change it. A gap between two sessions is a hypothesis to test with your logged context, not proof of either sensor error or physiology.
Step 3: Check what public validation actually covers
For each model you care about, fill in a coverage check. Inputs: model name, activities, reference. Judgment: does a study exist for that exact combination? Output: covered, partly covered or not covered.
A systematic review of commercial wearables measuring steps, energy expenditure and heart rate notes that questions remain about reliability and validity, especially for smaller and newer manufacturers where few or no studies exist (systematic review). That supports caution in general. It is not a per-model result.
| Question | Model with a published study | Model with no published study found |
|---|---|---|
| Public validation among the sources reviewed here? | Yes, for the specific device tested | None found |
| Activities covered | Only those in the protocol (for example rest, treadmill, resistance exercises) | Not established |
| Reference | Whatever the study used (chest strap, ECG) | Not established |
| Sample | The study's participants | Not established |
| Safe conclusion | Limited to that device, activity, reference and sample | No accuracy claim either way |
"No validation in the reviewed sources" is a coverage gap, not evidence of good or poor accuracy. Do not borrow one model's numbers to fill another's row, and do not carry a result from an older generation over to a newer watch in the same family.
Step 4: Build your comparison-limits table
An illustrative example (made-up numbers, not an actual test): you wear a watch on your left wrist for a 30-minute steady treadmill jog on Monday and a 30-minute rowing session on Thursday at the same logged effort. Monday's average reads 152 bpm and Thursday's reads 161 bpm. Your log shows less sleep before Thursday and a warmer room.
What you can say: the two sessions differ, and context could explain some of it. What you cannot say: whether the 9 bpm gap is optical artifact or real. To narrow it, repeat both with a chest strap worn at the same time and check whether the strap shows the same gap.
Use this table for any cross-model comparison:
| Check | What to record | What it limits |
|---|---|---|
| Model and generation | Exact name, firmware | Which studies apply |
| Wear position | Wrist, distance above bone, fit | Whether you match the study setup |
| Activity type | Steady cardio, intervals, lifting | Whether any evidence applies |
| Reference | Strap, ECG, or none | Whether you can judge error or only difference |
| Sample match | Age, skin tone, fitness level | How far a result generalizes |
| Metric and error measure | MAPE, correlation, bias | Correlation is not agreement; a device can track trends and still be off by a consistent margin |
| Data path | Watch, app, merged source | Which device produced each sample |
| Context log | Sleep, heat, caffeine, effort | Hypotheses only |
Accept your result when: every row is filled in, the compared sessions match on activity and reference, and your conclusion names the model, activity and reference. If a row is blank, say "not covered" instead of guessing.
The same limits apply to any wrist-worn sensor: log the setup, match the activity, and name the reference before drawing conclusions. If you want a wider look at zones once you trust the numbers, the heart rate zones guide is a sensible next read.
