top of page

10 WEARABLES · LAB-VALIDATED

Which wearable is actually accurate for heart rate?

At rest, your watch is nearly flawless. Then you move. The same wrist sensor that nails steady running can be 13x further off during racquet sport or weights. Here's how 10 devices really compare, and why arm motion, not intensity, is what breaks them.

5.5%

Best daytime error (Fitbit Charge 6)

1.67%

Best at night (Oura Gen 3, resting)

13x

Error swing, running vs badminton

10

Wearables tested, grouped in 4 tiers

Build your comparison · daytime

Compare brands side by side.

Pick 2–3 brands and every model we have daytime data for appears as its own column, on the five metrics that matter: overall error (MAPE), mean absolute error, bias, limits of agreement, and CCC agreement. All from the same daytime study against a chest strap (Gielen 2026). The better value in each row is highlighted, but small gaps sit inside the strap's own error, so we tell you the tier rather than crown a winner. Where a brand has two models, you will see how far apart they can be.

Choose brands to compare

Tap to add or remove · 2–3 brands · every model tested is shown

Fitbit

Garmin

Google

Apple

Polar

Xiaomi

Oura

MetricCharge 6Apple SEOura Gen 3Inspire 3
Overall error (MAPE) median % vs strap lower is better5.5%7.3%15.0%16.5%
Mean absolute error bpm off, median lower is better4.5 bpm5.0 bpm11.0 bpm14.3 bpm
Mean bias bpm, read alongside the spread context, not a win condition+0.7 bpm+0.9 bpm-7.1 bpm-14.4 bpm
Limits of agreement bpm spread, tighter is better lower is better23.9 bpm44.4 bpm56.8 bpm73.8 bpm
Agreement (CCC) concordance 0–1 higher is better0.930.700.610.45

These span Tiers 1–4 by day. Even so, this is one rotated protocol against a chest strap: every device here is far more accurate at rest and far worse during racquet sport, rowing or intervals. Read MAPE and the limits of agreement, not the bias.

Get the Kygo app

Accuracy by device · day vs night

These span Tiers 2–4 by day. Even so, this is one rotated protocol against a chest strap: every device here is far more accurate at rest and far worse during racquet sport, rowing or intervals. Read MAPE and the limits of agreement, not the bias.

MetricApple SEOura Gen 3
Overall error (MAPE) median % vs strap lower is better7.3%15.0%
Mean absolute error bpm off, median lower is better5.0 bpm11.0 bpm
Mean bias bpm, read alongside the spread context, not a win condition+0.9 bpm-7.1 bpm
Limits of agreement bpm spread, tighter is better lower is better44.4 bpm56.8 bpm
Agreement (CCC) concordance 0–1 higher is better0.700.61
MetricVivoactive 5Apple SEVivosmart 5Oura Gen 3
Overall error (MAPE) median % vs strap lower is better6.3%7.3%8.1%15.0%
Mean absolute error bpm off, median lower is better5.1 bpm5.0 bpm7.0 bpm11.0 bpm
Mean bias bpm, read alongside the spread context, not a win condition-1.0 bpm+0.9 bpm+4.8 bpm-7.1 bpm
Limits of agreement bpm spread, tighter is better lower is better35.3 bpm44.4 bpm41.2 bpm56.8 bpm
Agreement (CCC) concordance 0–1 higher is better0.830.700.780.61
MetricVivoactive 5Vivosmart 5Oura Gen 3
Overall error (MAPE) median % vs strap lower is better6.3%8.1%15.0%
Mean absolute error bpm off, median lower is better5.1 bpm7.0 bpm11.0 bpm
Mean bias bpm, read alongside the spread context, not a win condition-1.0 bpm+4.8 bpm-7.1 bpm
Limits of agreement bpm spread, tighter is better lower is better35.3 bpm41.2 bpm56.8 bpm
Agreement (CCC) concordance 0–1 higher is better0.830.780.61
MetricVivoactive 5Vivosmart 5Ignite 3Polar PacerOura Gen 3
Overall error (MAPE) median % vs strap lower is better6.3%8.1%11.2%13.1%15.0%
Mean absolute error bpm off, median lower is better5.1 bpm7.0 bpm9.5 bpm9.7 bpm11.0 bpm
Mean bias bpm, read alongside the spread context, not a win condition-1.0 bpm+4.8 bpm-4.3 bpm-3.9 bpm-7.1 bpm
Limits of agreement bpm spread, tighter is better lower is better35.3 bpm41.2 bpm54.3 bpm51.5 bpm56.8 bpm
Agreement (CCC) concordance 0–1 higher is better0.830.780.630.660.61

These span Tiers 1–3 by day. Even so, this is one rotated protocol against a chest strap: every device here is far more accurate at rest and far worse during racquet sport, rowing or intervals. Read MAPE and the limits of agreement, not the bias.

Great at night, grouped in tiers by day.

Use the toggle to switch the chart between daytime and nighttime accuracy. By day, the largest study (Gielen 2026) puts ten devices against a chest strap, but they were rotated (each tested on about ten people, not all 45) and the strap carries its own error, so we read it as four tiers, not a 1-to-10 list. At night, resting heart rate is the easy case and every validated device is excellent. Scroll sideways on mobile.

By day, moving

At night, resting

Mixed daytime movement vs a chest strap (Gielen 2026)

MetricVivoactive 5Vivosmart 5Ignite 3Polar Pacer
Overall error (MAPE) median % vs strap lower is better6.3%8.1%11.2%13.1%
Mean absolute error bpm off, median lower is better5.1 bpm7.0 bpm9.5 bpm9.7 bpm
Mean bias bpm, read alongside the spread context, not a win condition-1.0 bpm+4.8 bpm-4.3 bpm-3.9 bpm
Limits of agreement bpm spread, tighter is better lower is better35.3 bpm41.2 bpm54.3 bpm51.5 bpm
Agreement (CCC) concordance 0–1 higher is better0.830.780.630.66
WearableOverall error (MAPE)Agreement (CCC)Buy
Tier 1 Closest to the 5% line, but the gaps between these three are smaller than the strap's own error
Fitbit Charge 65.5%0.93Amazon
Garmin Vivoactive 56.3%0.83Amazon
Google Pixel Watch 26.7%0.87Amazon
Tier 2 Accurate on average, but individual readings swing wider
Apple Watch SE7.3%0.70Amazon
Garmin Vivosmart 58.1%0.78Amazon
Tier 3 Below the acceptability line on mixed daytime movement
Polar Ignite 311.2%0.63Amazon
Xiaomi Watch 211.9%0.69Amazon
Polar Pacer13.1%0.66Amazon
Tier 4 Bottom tier by day (the Oura ring is superb at night; see below)
Oura Ring Gen 315.0% 1.67% at night0.61Amazon
Fitbit Inspire 316.5% 5.4% in patients0.45Amazon

What was measured: each device against a Zephyr chest strap across one daytime protocol (seated rest, a stress task, treadmill and intermittent walking; Gielen 2026). MAPE is the average percent a reading is off, so lower is better; CCC is agreement with the reference from 0 to 1, where 1.00 is perfect. No device cleanly clears the <5% MAPE line.

Read the tier, not the rank. Devices were rotated: each was tested on about 10 of the 45 people, not all 45, and the chest strap carries its own ~2–4% error. Gaps smaller than that are noise, so we group into four tiers rather than a 1-to-10 list.

Bias can lie. A near-zero average bias just means over-reads and under-reads cancelled out. Read MAPE and the limits of agreement, never the single bias figure a brand quotes.

Model beats brand. The best and worst devices here are both Fitbits (5.5% and 16.5%). "Brand X is accurate" without a model number means nothing.

Get the free State of Wearable Accuracy report

What actually decides accuracy

Resting heart rate while asleep vs ECG (Dial 2025)

A different, smaller device set. Overnight validation exists only for these models (Dial 2025, 536 nights vs single-lead ECG); most watches in the daytime chart have simply never been tested at night, so we cannot re-rank the same ten here. Every device measured overnight lands near 1–3% error: the still body is the easy case, and it is where nearly every "99% accurate" marketing claim is actually taken. † Polar's figure covers the first four hours of sleep. Ranked on these figures, Oura is first, Polar second on 2.71% and WHOOP third on 3.00%.

Get the free State of Wearable Accuracy report

It is your arm, not the brand.

Accuracy is not decided by the brand or by how hard you work. It is decided by whether your arm is steady or jerking around. The same device is near-perfect at rest and during steady running, then falls apart the moment the wrist whips or flexes, which is why the tiers above are a mixed-daytime average, not a verdict on any one activity.

Steady & still

MAPE 1–5% best case

Rest, sleep, running

Rest, sleep and steady running are the easy conditions. Running is one of the easiest of all: a Fitbit hit 1.2% MAPE. At night, resting heart rate error drops to about 1.7 to 3%. This is where every "99% accurate" claim comes from.

Irregular arm movement

MAPE 13–17% where it breaks

Racquet, rowing, weights

Badminton, tennis, soccer, rowing and weightlifting whip or flex the wrist, so the optical sensor loses the pulse. The same Fitbit that ran at 1.2% hit 16.2% at badminton, under-reading by about 16 bpm, telling you that you worked less hard than you did.

The failure mode is irregular arm movement, not intensity. High-intensity running is one of the easiest cases; picking things up and racquet sport are the hard ones. Ceugniez 2025 · Vermunicht 2025 · Zhang 2020

Activity

Heart-rate error (MAPE) · lower is better

Running

steady, best case

1.2%

Cycling

some wrist flex

8.1%

Tennis

sharp arm rotation

8.9%

Orienteering

run + map handling

9.5%

Badminton

sharp arm whips

16.2%

Soccer

random arm actions

17.5%

These are error rates, not accuracy: the number is how far the watch is off (median MAPE), so a smaller bar is better. Fitbit Charge 4 vs a Polar H10 chest strap, same 26 people (Ceugniez 2025). Badminton produces about 13x the error of steady running on the same watch. During racquet and field sport the watch under-reads, telling you that you worked less hard than you did. Rowing lands the same way (13.4% MAPE, Vermunicht 2025).

Intervals are worse than all-out effort. Same bike, same people: error was 11.8% at maximal steady effort but jumped to 26.0% on intervals: the single worst number in the research. It is changing exercise that breaks the sensor, not hard exercise, which is why HIIT and stop-start sport read so poorly. Reddy 2018 · Merrigan 2023

The same sensor, two worlds

Great at night, shaky by day.

Wearables look their best when you are asleep, because a still body and steady blood flow are the easy case. That is also where nearly every manufacturer accuracy claim is measured, then quoted as if it held all day.

Oura Ring Gen 3

finger ring

At night

1.67%

By day

15.0%

About 9× the error once you move. Same finger sensor, median MAPE. Dial 2025 · Gielen 2026

Samsung Gear Sport

24h recording

At night

1.06 bpm

By day

11.10 bpm

About 10× worse the moment they woke up. Same device, one night, asleep vs awake (MAE). Sarhaddi 2022

This is the whole story in two devices. Same sensor, same person: just add movement, and the number falls apart. For the full list of what has actually been validated overnight, flip the chart above to At night, resting. Dial 2025 · Sarhaddi 2022

In detail

The full breakdown, tap any device.

Every wearable's independent finding versus a criterion standard, the manufacturer's claim, its error, bias and agreement numbers, and what it is best and weakest for.

Fitbit Charge 6

Independent finding vs a criterion standard

Top tier in the Gielen daytime protocol: median MAPE 5.5%, the tightest limits of agreement in the study (-11.2 to +12.7) and the only CCC above 0.90. It still does not clear the 5% acceptability line cleanly, and this is a daytime-motion figure, not a marketing sleep number. Devices were rotated (about 10 of the 45 people each), so read this as a tier, not a rank.

Error vs strap

MAPE

5.5%

MAE

4.5 bpm

Bias & agreement

Bias

+0.7 bpm

CCC

0.93

limits of agreement -11.2 to +12.7 bpm

Manufacturer claim

Google says a 2023 update made HR tracking "40% more accurate" for vigorous activity, but that claim attaches to the Pixel Watch 2, not the Charge 6, and cites no baseline, reference standard or study.

  • How it tracks Wrist band, optical PPG plus motion; Fitbit HR algorithm.
  • Best for Top-tier daytime accuracy in the Gielen protocol
  • Weakest for Still above the 5% line; racquet sport and rowing degrade it like any wrist PPG

View Fitbit Charge 6 on Amazon

Garmin Vivoactive 5

Independent finding vs a criterion standard

Top tier in the Gielen protocol (6.3% MAPE, CCC 0.83). Garmin is inconsistent rather than uniformly bad: strong here and in some resting data, weaker in others. It is also the only major brand that publishes no wrist-HR accuracy figure of its own.

Error vs strap

MAPE

6.3%

MAE

5.1 bpm

Bias & agreement

Bias

-1.0 bpm

CCC

0.83

limits of agreement -18.6 to +16.7 bpm

Manufacturer claim

Garmin publishes no numerical wrist-HR accuracy figure at all, alone among the major brands. Device manuals redirect to a general accuracy page with no number.

  • How it tracks Wrist watch, optical HR (Elevate) plus motion.
  • Best for A strong all-round daytime result and reliable steady-cardio tracking
  • Weakest for No published accuracy data to check a claim against; wide agreement limits

View Garmin Vivoactive 5 on Amazon

Google Pixel Watch 2

Independent finding vs a criterion standard

Top tier in the Gielen protocol (6.7% MAPE) with tight limits of agreement (-15.0 to +14.2) and CCC 0.87. Independently validated by Gielen 2026, which corrected earlier lists that had called it unvalidated.

Error vs strap

MAPE

6.7%

MAE

4.9 bpm

Bias & agreement

Bias

-0.4 bpm

CCC

0.87

limits of agreement -15.0 to +14.2 bpm

Manufacturer claim

Google's "40% more accurate for vigorous activity" claim is specifically for this device, but it gives no baseline, no reference standard and no supporting study.

  • How it tracks Wrist watch, multi-path optical HR sensor plus motion.
  • Best for Tight, well-centered readings across a mixed daytime protocol
  • Weakest for Vendor claim is unquantified; motion still degrades it like any watch

View Pixel Watch 2 on Amazon

Apple Watch SE

Independent finding vs a criterion standard

A textbook Rule 1 row: one of the smallest biases in the table (+0.9 bpm) sitting next to some of its widest limits of agreement (-21.4 to +23.0) and a CCC of only 0.70. The average looks excellent while individual readings swing hard. Note the tested device is the budget SE, not a flagship.

Error vs strap

MAPE

7.3%

MAE

5.0 bpm

Bias & agreement

Bias

+0.9 bpm

CCC

0.70

limits of agreement -21.4 to +23.0 bpm

Manufacturer claim

Apple's own white paper (Nov 2024, 100,000+ workouts) is the most transparent disclosure in the category: within 5 bpm 98% sedentary, 96% cycling, down to 87% walking. Its background algorithm reads 89% within 5 bpm on Series 6+ versus 72% on the SE and Series 4-5.

  • How it tracks Wrist watch, second-gen optical HR sensor plus motion.
  • Best for A tiny mean bias and strong transparency from Apple's own data
  • Weakest for Wide swings (low CCC 0.70); the flagship sensor was not the one tested

View Apple Watch SE on Amazon

Garmin Vivosmart 5

Independent finding vs a criterion standard

Second tier in the Gielen protocol (8.1% MAPE) but a +4.8 bpm over-read, the largest positive bias in the table. It posts strong correlations elsewhere (r 0.96 to 0.97 in Lee 2026 resistance testing), so the story is a consistent upward offset rather than noise.

Error vs strap

MAPE

8.1%

MAE

7.0 bpm

Bias & agreement

Bias

+4.8 bpm

CCC

0.78

limits of agreement -15.8 to +25.4 bpm

Manufacturer claim

As with all Garmin hardware, no numerical wrist-HR accuracy figure is published.

  • How it tracks Wrist band, optical HR (Elevate) plus motion.
  • Best for Consistent, predictable behavior (a steady over-read you can mentally adjust)
  • Weakest for Systematically reads about 5 bpm high; no vendor figure to verify

View Garmin Vivosmart 5 on Amazon

Polar Ignite 3

Independent finding vs a criterion standard

Third tier in the Gielen protocol (11.2% MAPE, CCC 0.63) and notably heat-sensitive: its error roughly doubled in a hot climate chamber (9.5% to 18.2%). Polar is unusual in publishing its own weak numbers for strength and team sport.

Error vs strap

MAPE

11.2%

MAE

9.5 bpm

Bias & agreement

Bias

-4.3 bpm

CCC

0.63

limits of agreement -31.4 to +22.9 bpm

Manufacturer claim

Polar is the only brand publishing its own bad numbers: overall MAE 4.4 bpm (3.7%), but strength training 5.8 bpm and floorball 12.6 bpm in its Precision Prime and Elixir white papers.

  • How it tracks Wrist watch, Polar Precision Prime optical sensor.
  • Best for An honest manufacturer that publishes its weak cases
  • Weakest for Below the acceptability line, and heat degrades it sharply

View Polar Ignite 3 on Amazon

Xiaomi Watch 2

Independent finding vs a criterion standard

Third tier in the Gielen protocol (11.9% MAPE) with some of the widest limits of agreement in the table (-33.3 to +27.2). Beyond this single study there is no independent or manufacturer HR validation to check it against.

Error vs strap

MAPE

11.9%

MAE

9.1 bpm

Bias & agreement

Bias

-3.0 bpm

CCC

0.69

limits of agreement -33.3 to +27.2 bpm

Manufacturer claim

No independent or manufacturer HR-accuracy figure is published for the Watch 2.

  • How it tracks Wrist watch, optical HR sensor plus motion.
  • Best for Budget option; acceptable at rest like most wrist PPG
  • Weakest for Wide swings under motion and no validation record

View Xiaomi Watch 2 on Amazon

Polar Pacer

Independent finding vs a criterion standard

Third tier in the Gielen protocol (13.1% MAPE, CCC 0.66), below the acceptability line on this mixed daytime protocol. Like the Ignite 3, it under-reads on average, so it tends to tell you that you worked less hard than you did.

Error vs strap

MAPE

13.1%

MAE

9.7 bpm

Bias & agreement

Bias

-3.9 bpm

CCC

0.66

limits of agreement -29.7 to +21.8 bpm

Manufacturer claim

Covered by Polar's Precision Prime white-paper figures (overall MAE 4.4 bpm), which are far better than this independent mixed-protocol result.

  • How it tracks Wrist watch, Polar Precision Prime optical sensor.
  • Best for Running-focused watch; steady cardio is its easy case
  • Weakest for Above the error line on mixed movement; tends to under-read

View Polar Pacer on Amazon

Oura Ring Gen 3

Independent finding vs a criterion standard

The split-personality device: excellent at night (1.67% MAPE against ECG, Dial 2025) but bottom tier by day (15.0% MAPE), because a finger sensor is not motion-artifact resistant. It is the clearest proof that "rings are more accurate" is marketing, not data: placement is not the variable that decides accuracy, motion is.

Error vs strap

MAPE

15.0%

MAE

11.0 bpm

also 1.67% at night

Bias & agreement

Bias

-7.1 bpm

CCC

0.61

limits of agreement -35.5 to +21.3 bpm

Manufacturer claim

Oura markets resting-HR r-squared of 0.996, but that traces to a study whose authors are all Oura-affiliated, and it is a nightly-average figure; for 5-minute segments it falls to about 0.87.

  • How it tracks Finger ring, optical PPG plus temperature.
  • Best for Nocturnal resting heart rate and the vitals measured while you sleep
  • Weakest for Daytime and any movement; bottom tier by day despite leading at night

View Oura Ring Gen 3 on Amazon

Fitbit Inspire 3

Independent finding vs a criterion standard

Bottom tier in the Gielen protocol (16.5% MAPE, under-reading by 14 bpm, CCC 0.45). But do not over-read it: the same device measured 5.40% MAPE against ECG in cardiac patients during exercise testing (Kitagaki 2025). Protocol and population, not the device alone, drive the number: the clearest illustration that model and protocol beat brand.

Error vs strap

MAPE

16.5%

MAE

14.3 bpm

also 5.4% in patients

Bias & agreement

Bias

-14.4 bpm

CCC

0.45

limits of agreement -51.3 to +22.5 bpm

Manufacturer claim

Google's 2023 "40% more accurate" line is often misapplied to the Inspire and Charge; it was specific to the Pixel Watch 2 and carries no study.

  • How it tracks Wrist band, optical PPG plus motion; Fitbit HR algorithm.
  • Best for A cheap tracker that is far more accurate at rest than this row suggests
  • Weakest for Worst mixed-motion result of the ten; large under-read

View Fitbit Inspire 3 on Amazon

The "View on Amazon" links above are affiliate links. As an Amazon Associate, Kygo Health earns from qualifying purchases, at no extra cost to you.

Make yours more accurate

Five ways to get a better reading.

You cannot change the sensor, but you control more than you think. These five levers move accuracy more than upgrading to a newer watch does, and four of them are free.

Biggest lever

Wear it higher up your arm

Moving a watch from one finger-width to three finger-widths above the wrist bone cut error by about 11 points and raised agreement from 0.59 to 0.92. The wrist bone is the worst possible spot for a sensor.

Vermunicht 2025

Free

Tighten the band (snug, not cutting off)

Contact pressure changes signal quality as much as, or more than, exercise intensity. A loose watch that slides lets in light and motion. Snug it up before a workout, then loosen it after.

Scardulla 2020

For hard sessions

Use an armband or strap for racquet & rowing

An optical band on the upper arm hit 1.35% error versus 6.82% on the wrist (same-brand comparison), and a chest strap is better still. For badminton, tennis, soccer, rowing and weights, an upper-arm band or chest strap beats any wrist device; the forearm moves far less than the wrist.

Schweizer 2025

Sensor contact

Keep the sensor on bare, unmarked skin

Tattoos directly under the sensor attenuate the signal, worst at rest. A shirt sleeve, a grimy sensor window or a gap over ink all cost you accuracy. Clean skin, direct contact.

Navalta 2025

Know its limits

Trust it for the right activities

Believe your wrist HR for running, walking, cycling and sleep, where error is a few percent. Distrust it for racquet sport, rowing and weightlifting, where it can be 13 to 17% off and usually reads low.

Ceugniez 2025 · Zhang 2020

True for every device

What no brand can fix.

Three things that hurt accuracy no matter which wearable you buy; they are about your body and your rhythm, not the sensor. Read these before you trust any single number a spec sheet gives you.

Irregular rhythm

Atrial fibrillation breaks it further

During peak exercise in AF the average error jumps to about 29 bpm, versus 14 bpm in normal rhythm, and devices under-read far more often than they over-read (61% vs 25% of readings). It is the rhythm during the reading that matters, not the diagnosis on your chart.

Quinn 2024 (12-lead ECG, N=81)

Skin tone

Fine at rest, off under load

Pooled data shows no significant pulse-rate bias by skin pigmentation at rest. Under exercise the picture changes: above ~60% effort, medium and dark skin ran about 11.8 bpm off while light skin stayed accurate. On current hardware (Apple Series 9) the limits of agreement widened 4.6-fold from the lightest to the darkest group. The fixes are the same for everyone.

Singh 2024 · Hung 2025 · Chase 2024

Tattoos

Ink over the sensor attenuates the signal

A tattoo directly under the sensor scatters the light the optics rely on, and the penalty is largest at rest, where the pulse signal is smallest; one study saw resting error climb from ~3% on clear skin to 23% over ink (much of it dropped readings). Movement masks it, rest exposes it. Keep the sensor on bare, unmarked skin.

Navalta 2025 (within-subject, N=25)

Claim vs reality

What the marketing says, and what the studies show.

Every big HR accuracy claim traces back to a study. When you read the study, the scope is almost always narrower than the ad. Here is each brand's headline claim next to what the research actually supports.

WHOOP

Misleading

“WHOOP is 99.7% accurate in measuring heart rate.”

Traces to one study that measured nocturnal HR only, one night, on the WHOOP 3.0. The scope is silently dropped, and the page is still live while WHOOP sells the 5.0 and MG. Against a real exercise ECG, WHOOP 4.0 was second worst of five.

WHOOP marketing page vs Miller 2022 / Van Oost 2025

Oura

Misleading

“Resting heart rate r-squared of 0.996.”

All four authors of the source study carry an Oura affiliation, and it is a nightly-average figure (an r-squared is not an accuracy percentage). For 5-minute segments the same relationship falls to about 0.87, and by day the Gen 3 is bottom tier.

Oura blog vs Kinnunen 2020

Samsung

Misleading

“30% more accurate; 90% correlation while running.”

The 30% baseline is Samsung’s own previous watch, so it is a relative claim with no absolute figure and no reference standard. The supporting university work is Samsung-commissioned, unpublished, with no model or sample size disclosed.

Samsung Newsroom

Google / Fitbit

Misleading

“40% more accurate for vigorous activity like HIIT and rowing.”

Applies to the Pixel Watch 2 specifically and is frequently misapplied to the Charge 6. There is no baseline, no reference standard and no study; the Fitbit support page lists no numerical HR figure at all.

Google blog 2023

Garmin

No figure

“None. Manuals redirect to a general accuracy page.”

Garmin publishes no numerical wrist-HR accuracy figure in its technology pages, manuals or support docs, alone among the major brands. Independent testing is the only way to judge it.

garmin.com/ataccuracy

Ultrahuman

Not validation

“Ring HR accurate to a mean error of 2.4–2.6 bpm.”

Based on n=6, and the comparison is against an Apple Watch and a SleepImage ring, not an ECG or chest strap. Agreement between two optical estimators is not validation against a truth source.

Ultrahuman self-published, Apr 2024

Apple

Transparent

“Within 5 bpm: 98% sedentary down to 87% walking; background 89% (Series 6+) vs 72% (SE).”

Nothing wrong. The most transparent disclosure in the category, across 100,000+ workouts, and the 72% background figure is the most useful number any brand publishes. Not peer reviewed, but honest about its own limits.

Apple white paper, Nov 2024

The bottom line

If you are buying for daytime tracking, the top tier is the Fitbit Charge 6, Garmin Vivoactive 5 and Google Pixel Watch 2, and the gaps between them are too small to call a winner. Oura Gen 3 is superb at night but bottom tier by day, because a finger sensor is not motion-proof. But the honest headline is not a device at all: at rest almost everything is accurate, and during racquet sport, rowing, weights or intervals almost everything is not. Pick for the job: a top-tier watch for steady cardio and sleep, and a chest strap or upper-arm band for stop-start sport. Read MAPE and the limits of agreement, not the bias a brand quotes, and treat any figure from a single night as a best case, not your day.

Keep exploring

Related tools.

More free, evidence-based tools to get the most out of your wearable.

Impact in bpm

Resting Heart Rate Factors

37 factors that move your resting heart rate, sorted by their exact impact in bpm.

Recovery · 37 factors

Open

What moves accuracy

Wearable Accuracy: What Actually Affects It

Tattoos, hairy arms, cold hands, strap tightness, which wrist. What the studies found and the free fixes that work tonight.

Wearables · 26 answers

Open

Fitbit Air vs WHOOP

Fitbit Air vs WHOOP

Heart-rate, sleep and calorie accuracy with numbers attached, plus the 3-year cost.

Wearables · head-to-head

Open

Want every brand on every metric in one place? Open the wearable accuracy hub

FAQ

Common questions.

Which wearable is the most accurate for heart rate?

Across three head-to-head studies combined, Apple Watch is first for exercise heart rate, Garmin second and Fitbit third. Cycling on the same riders, the Apple Watch Series 8 was off by 2.51 bpm, the Garmin Forerunner 955 by 2.93 bpm and the Fitbit Sense 2 by 3.89 bpm (Kostrna 2026). The largest daytime study (Gielen 2026) is read as tiers, not a ranking, because its devices were rotated across people: the top tier was the Fitbit Charge 6, Garmin Vivoactive 5 and Google Pixel Watch 2 (median MAPE 5.5 to 6.7%), with the Apple Watch SE at 7.3%. And "most accurate" depends entirely on what you are doing: at rest almost every device is within a few percent, while during racquet sport, rowing, weights or intervals even the best can be 13 to 26% off. No wrist wearable clears the 5% acceptability line cleanly across a full protocol.

Are wrist heart rate monitors accurate during exercise?

For steady exercise like running or walking, yes, often within 1 to 4% error. The problem is irregular arm movement. Badminton, tennis, soccer, rowing and weightlifting whip or flex the wrist, so the optical sensor loses your pulse and error climbs to 13 to 17%, usually under-reading. For those activities a chest strap is still the honest choice.

Why is my watch accurate at night but not during workouts?

Because night is the easy case: your body is still and blood flow is steady, so resting heart rate error drops to about 1.7 to 3%. Motion is what breaks optical sensors. This also matters because most manufacturer "99% accurate" style claims come from sleep or resting data and do not describe workout accuracy.

Is the average heart-rate number the one I should trust?

No. A near-zero average bias (Apple's pooled figure is -0.27 bpm) just means over-reads and under-reads cancelled out. The same data has limits of agreement of about -7 to +7 bpm. Look at MAPE (mean absolute percentage error) and the limits of agreement, which do not cancel, rather than the bias a brand quotes.

Does skin tone or a tattoo affect heart-rate accuracy?

At rest, pooled data shows no significant bias by skin pigmentation. Under hard exercise the picture changes: one study found medium and dark skin about 11.8 bpm off above 60% effort while light skin stayed accurate. Tattoos directly over the sensor attenuate the signal, worst at rest. The fixes are the same for everyone: wear it snug, higher on the arm, on unmarked skin.

Are the newest watches (Apple Series 11, Galaxy Watch 7, WHOOP 5) more accurate?

Unknown independently. Peer-reviewed HR validation lags new hardware by about 2 to 4 years, so most current flagships carry manufacturer-only claims. The devices with independent testing are mostly a generation or two back, and firmware updates change results even on identical hardware.

Sources

Every claim, traceable.

Each figure checked against the primary record (journal, PMC or manufacturer paper). 29 sources. Verified August 2026.

Independent · 10 devices

Gielen et al. 2026: 10 wearables vs chest strap

JMIR Formative Res. 2026;10:e85186. N=45, 3 climates.

Independent · 12-lead ECG

Van Oost et al. 2025: 5 devices vs 12-lead ECG

Sensors. 2025;25(20):6319. N=24, missingness disclosed.

Independent · cycling

Kostrna et al. 2026: 4 watches on a 10-minute cycle

PLOS ONE. 2026;21(7):e0353261. N=58. MAE: Apple Watch Series 8 2.51, Garmin Forerunner 955 2.93, Fitbit Sense 2 3.89 bpm.

Independent · nocturnal

Dial et al. 2025: Oura, Polar & WHOOP overnight

Physiological Reports. 2025;13:e70527. 536 nights (AFRL).

Independent · nocturnal

Miller et al. 2022: sleep/wake HR across 6 devices

Sensors. 2022;22(16):6317. N=53, II-lead ECG.

Independent · meta

Zhang et al. 2020: HR by activity, 15 brands

J Sports Sci. 2020;38(17):2021. 44 articles, 738 effects.

Independent · by activity

Ceugniez et al. 2025: Fitbit by sport (run vs badminton)

JMIR mHealth uHealth. 2025;13:e54871. N=26, Polar H10.

Independent · rowing & position

Vermunicht et al. 2025: wear position & rowing

Eur Heart J Digital Health. 2025;6(5):1024. N=50.

Independent · intervals

Reddy et al. 2018: intervals vs maximal effort

JMIR mHealth uHealth. 2018;6(12):e10338. N=20, Polar H7.

Independent · intervals

Merrigan et al. 2023: rucking, cycling & circuits

Meas Phys Educ Exerc Sci. 2023;27(3):234. N=8, Holter ECG.

Independent · day vs night

Sarhaddi et al. 2022: 24-hour sleep vs wake split

PLOS ONE. 2022;17(12):e0268361. N=28, Shimmer3 ECG.

Independent · resistance

Lee et al. 2026: endurance vs resistance, 4 watches

Sensors. 2026;26(8):2526. N=62 men, ECG (SOLUM-funded).

Independent · Apple meta

Lambe et al. 2026: Apple Watch living meta-analysis

npj Digital Medicine. 2026;9:63. HR 22 studies, n=1,247.

Independent · Fitbit meta

Chevance et al. 2022: Fitbit meta-analysis

JMIR mHealth uHealth. 2022;10(4):e35626. 32 HR studies.

Independent · review

Fuller et al. 2020: within-tolerance across brands

JMIR mHealth uHealth. 2020;8(9):e18694. 177 comparisons.

Independent · atrial fibrillation

Quinn et al. 2024: AF vs sinus rhythm on exertion

J Am Coll Cardiol. 2024;83(12):1177. N=81, 12-lead ECG.

Independent · ballistic

Moghaddam et al. 2026: placement during burpees

Sensors. 2026;26(1):176. N=28, Polar H10.

Independent · tattoos

Navalta et al. 2025: PPG over tattooed skin

Sensors. 2025;25(22):6896. N=25, within-subject.

Independent · skin tone

Hung et al. 2025: skin tone by exercise intensity

PLOS ONE. 2025;20(2):e0318724. N=25, Fitbit vs Polar H10.

Independent · skin tone

Chase et al. 2024: skin tone on current hardware

J Funct Morphol Kinesiol. 2024;9(4):275. Apple Series 9, N=30.

Independent · pigmentation meta

Singh et al. 2024: skin pigmentation meta-analysis

J Med Internet Res. 2024;26:e62769. Pulse-rate arm n=176.

Independent · contradiction

Kitagaki et al. 2025: Fitbit Inspire 3 in patients

JMIR Cardio. 2025;9:e77911. N=30, ECG, MAPE 5.40%.

Manufacturer

Apple: Heart rate on Apple Watch white paper

Apple Inc. Nov 2024. 100,000+ workout sessions.

Manufacturer

WHOOP: 99.7% accuracy claim page

Traces to Miller 2022 (nocturnal, one night, WHOOP 3.0).

Manufacturer

Oura: resting-HR r-squared 0.996 claim

Kinnunen et al. 2020, all authors Oura-affiliated.

Independent · wear position

Schweizer & Gilgen-Ammann 2025: upper arm vs wrist

JMIR Cardio. 2025;9:e67110. Upper-arm 1.35% vs wrist 6.82%.

Independent · contact pressure

Scardulla et al. 2020: sensor contact pressure & PPG

Sensors. 2020;20(18):5052. N=17 (research prototype).

Manufacturer

Samsung: "30% more accurate" Galaxy Watch claim

Samsung Newsroom, Jul & Sep 2024. Baseline is prior watch.

Manufacturer

Google: Pixel Watch 2 "40% more accurate" claim

Google blog, Oct 2023. No baseline or study cited.

Show all 29 sources

From the blog

Show fewer sources

Keep reading.

The long-form, evidence-based articles behind this tool.

Wearables & Data

How Accurate Is Your Heart Rate Monitor? 10 Devices Tested

Ten wearables, the same 45 wrists, the same day, and error running from 5.5% to 16.5%. Which device you own matters less than you think.

12 min read

Read

HRV & Recovery

Heart Rate Accuracy by Activity Type: What Breaks Your Watch

1.2% error during a run and 16.2% during badminton, on the same watch. What breaks wrist heart rate is not how hard you work.

13 min read

Read

HRV & Recovery

Why Is My Resting Heart Rate Suddenly Higher? A Data-Driven Breakdown

Your resting heart rate jumped overnight and your wearable cannot say why. The ten most common causes, from late meals to overtraining.

7 min read

Read

This content is for informational purposes only and is not medical advice. Optical wrist and ring heart rate is an estimate, not a clinical measurement, and can be inaccurate during exercise or with an irregular heart rhythm. Consult a qualified healthcare provider before making health decisions based on wearable data.

Data from peer-reviewed validation studies and manufacturer documentation. Last updated October 25, 2026.

As an Amazon Associate, Kygo Health earns from qualifying purchases. Product links on this page are affiliate links; we may earn a commission at no extra cost to you.

New York, NY​

© 2025 by KYGO Health LLC Kygo Health LLC is not intended to diagnose, treat, cure, or prevent any disease. The information provided is for educational purposes only and is not a substitute for professional medical advice. Consult your physician before making any health decisions.

bottom of page
Text summary of this page (tap to expand)

About this heart rate accuracy comparison

This free tool compares 10 wearables on how accurately they measure heart rate against ECG and chest strap reference, using median MAPE, mean absolute error, bias, limits of agreement, and concordance correlation coefficient. Across three head to head studies combined, Apple Watch is first for exercise heart rate, Garmin second and Fitbit third: cycling on the same riders, the Apple Watch Series 8 was off by 2.51 bpm, the Garmin Forerunner 955 by 2.93 bpm and the Fitbit Sense 2 by 3.89 bpm. The largest 10 device test rotated watches across participants, so it is read as tiers rather than a ranking: the Fitbit Charge 6 (5.5 percent), Garmin Vivoactive 5 (6.3 percent) and Google Pixel Watch 2 (6.7 percent) formed the top tier, with the Apple Watch SE at 7.3 percent. The decisive variable is not the brand, it is arm movement: at rest almost every device lands within a few percent, while during racquet sport, rowing, or weights even the best wrist devices run 13 to 17 percent off. No wrist wearable clears the 5 percent acceptability line cleanly across all activities. Manufacturer funded validation is labeled separately from independent research. No signup required.

It is built by Kygo Health. Kygo Health is an iOS and Android app that connects AI food logging to wearable data and shows how what you eat affects your sleep, HRV, resting heart rate, energy, and recovery. It syncs Oura Ring, Apple Health, Health Connect, Fitbit, Garmin, and WHOOP. Food logging, wearable syncing, and nutrition tracking are free, including your first 5 AI photo logs. Pro, which unlocks the correlation engine, costs $9.99 per month, $49.99 per year with a 7 day free trial, or $99.99 once for lifetime access. Kygo is a general wellness product, not a medical device.

WearableNocturnal error (MAPE)Agreement (CCC)
All excellent Still body, steady blood flow: at night there is really only one tier
Oura Ring Gen 31.67%0.97
Oura Ring Gen 41.94%0.98
Polar Grit X Pro†2.71%0.86
WHOOP 4.03.00%0.91