Wearable heart rate accuracy dataset
The library includes 223 studies. The combined table uses device-level numbers from 139 of them. Read the report.
Heart rate accuracy by device
1,270 device results from 139 studies.
Error shows how far off the device was. Reads high or low shows whether its readings averaged above or below the reference.
| 1 | 0.67 bpm | 1Preliminary | Near even−0.18 bpm1 study | |
| 2 | 0.78 bpm | 1Preliminary | Low−1.52 bpm2 studies | |
| 3 | 0.9 bpm | 1Preliminary | Near even+0.1 bpm1 study | |
| 4 | 0.92 bpm | 2 | Near even−0.29 bpm2 studies | |
| 5 | 0.99 bpm | 1Preliminary | Low−1.34 bpm2 studies | |
| 6 | 1.18 bpm | 1Preliminary | Near even−0.31 bpm1 study | |
| 7 | Generation unspecified · Jeong Y et al. (2026) | 1.78 bpm | 1Preliminary | Near even+0.05 bpm1 study |
| 8 | 1.81 bpm | 1Preliminary | Near even+0.14 bpm4 studies | |
| 9 | 1.9 bpm | 1Preliminary | Low−0.99 bpm2 studies | |
| 10 | 1.94 bpm | 1Preliminary | Low−0.61 bpm1 study | |
| 11 | 2.08 bpm | 1Preliminary | Near even−0.03 bpm1 study | |
| 12 | 2.6 bpm | 1Preliminary | Low−2.1 bpm1 study |
Lower error ranks higher for the selected activity and error measure. A one-study rank is preliminary.
How the studies are combined
We extracted absolute errors from 98 studies. Each study contributes one estimate per device. Treadmill results include supervised clinical tests. We use the whole test when available, or the median across its speed or intensity stages.
We average the study estimates. The weight is the square root of the smallest reported participant count for those tests, capped at 100 participants. We use 10 participants as a weighting default when the count is unavailable. Manufacturer white papers and studies funded by a tested device maker receive half weight. With one study, that study supplies the whole estimate. Open a row to see each source’s share, or switch off weighting to give each study equal weight.
Direction summarizes signed differences in bpm against the reference. A negative value means the wearable read lower on average. Mixed means study estimates fell on both sides of zero outside a 0.5 bpm band. Near even means all the estimates fell within that band. The Studies column counts accuracy sources. Direction shows its own source count.
The ranks exclude one-person reports and results optimized by retrospectively shifting device timestamps. When a study gives only a family name or pools model variants, the row represents that study’s tested device or group. Protocols, wear sites, sampling windows and people vary. Open a row to see its original tests.
Find a study
The library covers 223 studies. Vora reanalyzed 7,518,782 paired readings from 42 people in two public datasets.
Each study has a heart-rate finding, the tested devices, the reference measurement and a link to the original work. Studies use different activities and statistics. Compare devices tested together under the same conditions.
Showing 12 of 223 studies
Apple Watch Heart Rate Accuracy Study
Apple reported a lower error for Series 12 in 139 of 159 activity-by-measure comparisons at its 99% confidence criterion. The paper used Apple’s internal logger for Series 12 and device-specific export streams for rivals.
- Devices
- Apple Watch Series 12 and six comparator devices
- Reference
- Polar H10 ECG chest strap
- Study sample
- 1,254 people
- Where to look
- Figures 1–3
Impact of Helicopter Vibrations on In-Ear PPG Monitoring for Vital Signs-Mountain Rescue Technology Study (MoReTech)
Pulse-rate concordance with ECG was 0.96 both at rest and under helicopter-simulator vibration.
- Devices
- In-ear optical prototype
- Reference
- Clinical ECG
- Study sample
- 17 people
- Where to look
- Abstract
A Speed-Dependent Assessment of E-Textile-Based Sensor Technology: Validity of the Prevayl Wearable Heart Rate Monitor
In a 16-minute incremental treadmill test, median beat count from the textile sensor was 100.5% of the reference. The study also compared heart rate at different speeds.
- Devices
- Prevayl smart textile monitor
- Reference
- 3-lead ECG
- Study sample
- 26 people
- Where to look
- Abstract
Pulse vs. R-R Interval: Validity and Reliability of the Garmin Elevate v4 Sensor for Ultra-Short HR and HRV Monitoring in Healthy Adults.
During short seated and supine tests, resting heart rate correlated at 0.94 or higher against ECG.
- Devices
- Garmin Forerunner 955
- Reference
- BIOPAC ECG
- Study sample
- 23 people
- Where to look
- Section 3.4 and Figure 6
Validity of the Polar H10 for Continuous Measures of Heart Rate and Heart Rate Synchrony Analysis.
Polar H10 heart rate correlated above 0.99 with ECG during seated audiovisual tasks.
- Devices
- Polar H10
- Reference
- Laboratory ECG
- Where to look
- Abstract
Accuracy of heart rate measurement using AirPods Pro 3 during graded treadmill exercise: A laboratory-based validation study.
Mean absolute error was 2.08 bpm in five-second treadmill windows. The agreement limits were roughly −10 to +10 bpm, with more spread at higher rates.
- Devices
- AirPods Pro 3
- Reference
- Polar H10
- Study sample
- 40 people
- Where to look
- Abstract and Table 2
Accuracy of Optical Heart Rate Measurements for 10 Commercial Wearables in Different Climate Conditions and Activities: Instrument Validation Study.
Across the climate chamber protocol, median participant mean absolute error was 4.5 bpm for Fitbit Charge 6 and 4.9 bpm for Pixel Watch 2.
- Devices
- Fitbit Charge 6, Fitbit Inspire 3, Garmin Vivosmart 5, Garmin Vivoactive 5, Apple Watch SE, Google Pixel Watch 2, Polar Ignite 3, Polar Pacer, Xiaomi Watch 2, Oura Ring Gen 3
- Reference
- Zephyr BioHarness ECG
- Study sample
- 45 people
- Where to look
- Tables 1–3
Heart Rate Estimation Using the Galaxy Watch During Maximal Cardiopulmonary Exercise Testing: Cross-Sectional Validation Study.
During maximal treadmill testing, the mean bias was −2.67 bpm, with limits of agreement from −16.90 to +11.57 bpm.
- Devices
- Samsung Galaxy Watch 6
- Reference
- Polar H10
- Study sample
- 55 people
- Where to look
- Abstract
Comparison of heart rate measurement accuracy among commercially available photoplethysmography-based wearable devices across exercise intensities.
Errors generally fell with intensity during this treadmill protocol. Apple Watch SE and Polar Verity Sense had lower mean absolute errors overall than the two other devices.
- Devices
- Polar Verity Sense, Polar Pacer Pro, Apple Watch SE, Galaxy Watch 6
- Reference
- Polar H10
- Study sample
- 37 people
- Where to look
- Table 3 and Figure 4
Validation of a novel multi-sensor wearable for measuring heart rate, steps, and energy expenditure in laboratory and simulated free-living protocols
Absolute percentage error ranged from 0.61% to 5.18% at rest and in recovery, and from 32.96% to 41.77% while jogging, depending on wrist placement.
- Devices
- DOFIT PRO
- Reference
- ECG chest strap
- Where to look
- Abstract
Smart swim goggles accurately measure heart rate during swim training: A criterion validation study.
Mean workout heart-rate error averaged 1.0% in the swim comparison. Concordance was 0.99 for workout averages and 0.91 for intervals.
- Devices
- Optical swim goggles
- Reference
- Garmin HRM-Swim
- Study sample
- 25 people
- Where to look
- Abstract
PPG-Based Heart Rate Accuracy in Hispanic Adults with Fitzpatrick III-V Skin Tones: An Evaluation of Body Composition and Skin-Tone Effects.
During cycling, absolute error averaged 2.51 bpm for Apple and 2.93 bpm for Garmin. The paper reports different usable sample sizes by model.
- Devices
- Apple Watch Series 8, Garmin Forerunner 955, Fitbit Sense 2, Samsung Galaxy Watch 5
- Reference
- Polar H10
- Study sample
- 58 people
- Where to look
- Table 1
How the studies were selected
We searched PubMed and PubMed Central via Europe PMC for human studies comparing heart rate from a wearable against ECG, ECG-derived polysomnography, or an ECG chest strap. The expanded search returned 1,663 distinct records to screen. We followed references from published reviews and included two labeled manufacturer reports with directly reported test results. We used the original abstract or the named table or figure to check each finding. Reviews, sensor simulations, tests of rhythm detection alone, and known duplicate reports of the same experiment were excluded from the count.
A study can include several devices and activities. The library shows a selected finding per study and retains its original metric. Participant counts describe the study or the sample for the shown result when available. The two public reanalysis cohorts are included in the study count. Other studies can have data available only in an abstract or a published table.
Download the data
The study library has one row per study. The combined dataset has one row per model, condition and study. Each keeps its own reference and sample. Read the analysis methods.
- Combined device results (CSV)
Device-by-condition results across studies, with separate error measures, sample counts and exact source locations.
- Combined device results (JSON)
Machine-readable device results across studies.
- Study library (CSV)
One row per primary study, with devices, reference, setting, sample, reported result and a citation.
- Study library (JSON)
Machine-readable version of the full study library, with DOI and PubMed IDs where available.
- Initial set of device results (CSV)
The first 301 records, with available error in bpm and percent, bias, spread, uncertainty and coverage.
- Initial set of device results (JSON)
Machine-readable version of the first 301 device results.
- Detailed source notes
Reference instruments, populations, funding, study limits and precise table or figure locations for the device results and context.
- Detailed source notes (JSON)
Machine-readable version of the detailed source notes.
- Treadmill sensitivity checks
Within-person changes in error and participant bootstrap intervals at three rapid-change cutoffs.
- Full treadmill reanalysis (JSON)
Condition-level aggregate results, participant counts and paired estimates for every cutoff.
- Treadmill analysis code
Reproduce the results from the original public ECG and wearable recordings.
- Free-living and swimming reanalysis (JSON)
Participant-weighted summaries from both public cohorts and the treadmill heart-rate matching results.
- Heart-rate-matched treadmill comparisons
Within-person changes with and without matching, with participant spread and uncertainty.
- Error at different reference rates
Typical absolute error, bias and top-end misses at different reference heart rates within each activity.
- Expanded analysis code
Reproduce the activity summaries and the heart-rate-matched treadmill comparison from the public recordings.
Cite as Vora Team, Wearable heart rate accuracy dataset, version 11, . When quoting a published value, credit the original paper. The two original datasets are released under CC BY 4.0. Free-living and lab recordings · Swimming recordings
Detailed source notes
- Yun, Ko & Lee · Resistance exercise
Peer-reviewed · 2026 · 70 healthy young men
Reference: ECG display. Where to look: Tables 1 and 3.
Four devices worn during eight sets. The authors compared displayed heart rate before each set, immediately after it, and one minute later.
Fixed exercise order. Spot readings reflect the displayed number, including any sensor, processing or display delay. The sample was young men. No external funding or conflicts reported.
- Vora · Treadmill reanalysis
Public data reanalysis · 2024 · 32 adults in the source study, with device-specific paired subsets
Reference: Medical-grade ECG. Where to look: Treadmill-running rows in the authors’ released 1 Hz files.
Vora recalculated absolute error from the aligned ECG and sensor readings, then compared steadier and rapidly changing heart rate within each person.
Exploratory cutoffs chosen after a pilot. The authors had already aligned device and ECG signals. Each device has its own eligible subset of people. The authors reported no conflicts. The analysis uses the public CC BY 4.0 release.
- Gielen et al. · Walking, stress and climate
Peer-reviewed · 2026 · 45 adults across the study, 10 test assignments per model
Reference: Zephyr BioHarness ECG chest strap. Where to look: Tables 1–3.
Ten devices were compared against an ECG chest strap in a climate chamber, with steady and stop-start walking and a seated stress task. The tables give medians and middle-50% ranges of participant errors.
Participants wore a subset of devices. Activity differences varied by model. The study did not detect a climate effect after adjustment for multiple testing. Internal KU Leuven funding. The funder reported no role in the conduct or interpretation.
- Temple · Treadmill zones
Manufacturer white paper · 2026 · 43 people, 42 with Fitbit Air recordings
Reference: Taped Polar H10 ECG chest strap. Where to look: Tables 2–3 and Figure 3, pages 4, 9–10.
Temple compared its head-worn tracker with five consumer devices during a graded treadmill protocol. Published errors average the participants’ mean absolute errors.
The vendor tested its own device. Its main error estimates exclude defined failure episodes and missing readings. Devices logged at different frequencies. Conducted and published by Temple, maker of one tested sensor.
- Van Oost et al. · Activity transitions
Peer-reviewed · 2025 · 24 adults, 23 with usable 12-lead ECG
Reference: 12-lead ECG. Where to look: Table 3 and Figure 6.
A treadmill protocol compared steady states with changes in posture, speed, incline and stopping. Fitbit Charge 5 and Sense 2 showed higher errors during transitions in the 10-second comparison.
Healthy university students. Shorter windows paired fewer readings for some devices. Missingness is measured relative to available ECG observations and depends on the window. Funded by Flanders Innovation & Entrepreneurship. No conflicts reported.
- Apple · Workout and daily living
Manufacturer white paper · 2026 · 1,254 people contributed at least one paired comparison
Reference: Polar H10 ECG chest strap. Where to look: Figures 1–3.
Series 12 was compared with six devices during activities including outdoor running, cycling, HIIT, strength and daily living. Figures report the Apple-minus-comparator difference for each outcome.
Apple used its own internal five-second logger and different export streams for rivals. The paper reports paired differences, which cannot fill an absolute-error column. Oura was excluded from lifting. Designed and published by Apple, which ran two sites. A contract research organization ran three sites.
- Gillinov et al. · Aerobic exercise
Peer-reviewed · 2017 · 50 healthy adults
Reference: ECG and Polar H7 chest strap. Where to look: Abstract and results by exercise.
Historical device comparisons at rest and during treadmill, bike and elliptical exercise. Agreement varied by exercise modality.
Older devices and concordance statistics. Useful for the effect of activity, with model-specific results limited to the hardware studied. See the paper for author disclosures.
- Dial et al. · Overnight heart rate
Peer-reviewed · 2025 · 13 healthy adults tested across repeated nights
Reference: Polar H10 single-lead ECG. Where to look: Nightly resting heart rate analyses.
Oura Gen 3 and Gen 4, Polar Grit X Pro and WHOOP 4.0 were compared with overnight ECG for resting heart rate. HRV was evaluated separately.
Repeated nights came from a small group. Nightly resting heart rate and live workout heart rate use different sampling windows. Air Force Research Laboratory funding. No competing interests reported.
- Vora · Free-living and lab recordings
Public data reanalysis · 2024 · 32 adults in the source study, with device-specific paired subsets
Reference: Medical-grade ECG. Where to look: Free-living, treadmill and day-one lab activity rows in the 1 Hz files.
Vora recalculated the size and direction of errors, including the spread between people, from the authors’ aligned ECG and device recordings.
Scosche Rhythm24 has no free-living data in the release. Laboratory cycling and daily tasks use day one. Counts refer to usable pairs in the released files. The authors reported no conflicts. The analysis uses the public CC BY 4.0 release.
- Vora · Running and swimming recordings
Public data reanalysis · 2026 · 10 trained swimmers
Reference: Polar H10 ECG chest strap. Where to look: Zenodo CSV, split by activity; device models in the source paper.
The same people wore both watches on land and in the pool. The public file includes only times when all three devices returned a value.
The chest strap may also be affected underwater. These results describe agreement with the strap during valid pairs. The release cannot tell us how often readings were missing. The original study reported no external funding or conflicts. The CSV is released under CC BY 4.0.
- Lee, Bae & Ko · Lifting across three visits
Peer-reviewed · 2026 · 60 healthy men, across three sessions
Reference: ECG display. Where to look: Tables 2, 3 and 5.
The authors tested a separate cohort from the 70-person lifting experiment. They recorded displays before the set and 5, 30 and 60 seconds after it, at all three visits.
The 5-second agreement limits are for individual observations and are marginal across the reference heart rates. The authors found proportional bias for all four devices. No external funding or conflicts reported.
- Budig et al. · Outdoor exercise and swimming
Peer-reviewed · 2021 · 36 endurance-trained adults
Reference: Polar H10 ECG chest strap. Where to look: Table 1.
The paper reports absolute differences and a percentage measure that divides by the mean reference heart rate.
Older device models and firmware. The authors call for further testing of the reference strap in water. No external funding or conflicts reported.
- Navalta et al. · Trail running
Peer-reviewed · 2020 · 21 healthy adults
Reference: Polar H7 ECG chest strap. Where to look: Tables 1–3.
Participants ran uphill and back downhill at a self-selected pace. The Suunto entry uses its accompanying ECG strap. Values are transcribed from the tables.
Older device models. The authors used different export pipelines for some devices. Device-specific counts and agreement limits are omitted from this extraction. No specific funding or conflicts reported.
- Nissen et al. · Everyday tasks
Peer-reviewed · 2022 · 23 healthy adults enrolled, 20 contributed analyzable readings
Reference: Mind Media NeXus-10 MKI Holter ECG. Where to look: Table 4.
Participants wore both devices for ten short tasks, from sitting and typing to walking and squats.
ECG recording issues reduced the available sample for some tasks. Two inconsistent limits in the published table are omitted. No conflicts declared.
- Kostrna et al. · Stationary cycling
Peer-reviewed · 2026 · 58 Hispanic adults enrolled, with 33 to 36 usable participants per tested model
Reference: Polar H10 ECG chest strap. Where to look: Table 1.
Participants held the cycle handles to limit wrist motion. The table compares the four consumer models with usable data.
This controlled cycling protocol does not capture the wrist motion of outdoor sports. The usable sample varies by model. NSF PATHS-UP pilot grant. The authors declared no competing interests.
- Ceugniez et al. · Six field sports
Peer-reviewed · 2025 · 26 young adults across 55 sessions
Reference: Polar H10 ECG chest strap. Where to look: Tables 2 and 3.
The authors aligned and smoothed the time series over ten seconds before computing error.
The participants and number of sessions differ by sport. Cycling and tennis each had just two participants. No conflicts declared.
- Miller et al. · Overnight and presleep heart rate
Peer-reviewed · 2022 · 53 healthy adults
Reference: Laboratory ECG. Where to look: Table 6 and Figure 2.
Absolute error is the average size of the difference in summary heart rate. Devices use different windows, including presleep sampling for Polar.
One laboratory night. These device generations have since been succeeded by newer hardware. Funded by the Australian Institute of Sport. The authors’ lab receives separate support from WHOOP, which had no role in this study.
- Danielsson et al. · Wheelchair treadmill propulsion
Peer-reviewed · 2024 · 20 manual wheelchair users and 20 people without a disability
Reference: Polar H10 ECG chest strap. Where to look: Results, Absolute Agreement.
The authors combined three incline settings and three stages of increasing speed within each group to estimate mean absolute percentage error.
The Apple Watch used its wheelchair setting. The Fitbit used treadmill running mode. See the original study’s disclosures.
- Navalta et al. · Walking, jogging and skipping
Peer-reviewed · 2023 · 18 adults
Reference: Polar H10 ECG chest strap. Where to look: Table 3.
Participants walked, jogged and skipped at self-selected paces in five-minute bouts. Published biases and agreement limits are expressed here as wearable minus reference.
The published jogging pair count for one model is inconsistent. Pair counts are omitted. See the original study’s disclosures.