How Accurate Is WHOOP? HR, HRV and ECG Evidence

Overnight heart rate from WHOOP stays close to an ECG reference in the sleep studies below. Overnight HRV, reported as RMSSD, follows the reference on average and has a wider error, so it is a personal trend rather than a one-night clinical value. WHOOP MG adds a separate on-demand ECG: a 30-second recording the FDA has cleared to label atrial fibrillation versus sinus rhythm. That ECG is not the sensor behind the nightly heart-rate or HRV numbers.

The rest of this page walks through hardware, exercise heart rate, sleep, strain, recovery, and calories. For the optical method itself, see the complete guide to PPG technology.

WHOOP 4.0 and 5.0 Sensor Hardware

The PPG Sensor System

WHOOP 4.0 uses a reflectance-mode PPG sensor array on the underside of the strap, positioned against the dorsal surface of the wrist by default. The sensor includes:

• Green LEDs (approximately 525 nm) for heart rate measurement
• Red and infrared LEDs for SpO2 measurement
• Photodiodes for detecting reflected light
• 3-axis accelerometer for motion detection and activity classification
• Skin temperature sensor for metabolic and recovery insights
• Skin conductivity sensor as an additional data stream

WHOOP 5.0 iterates on this sensor package with improved LED efficiency, additional photodetectors for better spatial coverage, and enhanced signal processing capabilities. The physical form factor remains a strap design that can be worn on the wrist or in a body strap at the bicep, forearm, or other body locations.

The strap design differentiates WHOOP from rigid-cased watches like Apple Watch or Garmin. The flexible strap can conform more closely to the wrist contour, potentially improving optical coupling compared to a flat watch caseback. However, the strap can also shift more easily during vigorous movement, which introduces its own motion artifact challenges. For more on how different wearable form factors affect PPG accuracy, see our PPG sensor placement guide.

Heart Rate Accuracy

During sleep, published comparisons put WHOOP heart rate within about 1–2 bpm of the reference on average. The trials below used WHOOP 2.0, 3.0, and 4.0. They are earlier hardware than WHOOP 5.0 and the MG strap. Dial et al. (2025) compared WHOOP 4.0 nocturnal resting heart rate with a Polar H10 chest-strap ECG in 13 healthy adults (age 33.2 ± 8.6 years) at home. Across 288 WHOOP nights, mean bias was −1.41 ± 1.69 bpm, limits of agreement −4.72 to 1.90 bpm, mean absolute error 1.78 ± 1.31 bpm, MAPE 3.00 ± 2.15%, and Lin's concordance 0.91. The authors' scale calls that concordance moderate, and their conclusion calls resting heart rate acceptable for WHOOP 4.0. The chest-strap value was an average over the whole night. WHOOP's reported resting heart rate is a dynamic sleep average weighted toward the last slow-wave sleep stage, so part of the difference is the window.

Miller et al. (2022) compared WHOOP 3.0 with a laboratory ECG in 53 adults (age 25.4 ± 5.9 years) on one night in a sleep laboratory. Heart rate, calculated by the researchers from R-R intervals WHOOP supplied for the whole sleep period, had bias −0.3 ± 1.0 bpm, absolute bias 0.7 ± 0.8 bpm, limits of agreement ±1.9 bpm, and ICC 0.99. Bellenger et al. (2021) compared WHOOP 2.0 with ECG during slow-wave sleep in 6 adults across 15 sleep opportunities. Heart-rate bias was ≤0.39 ± 0.38% and limits of agreement were ≤1.56%, lower than the day-to-day coefficient of variation they cite for heart rate (10–11%). Bias in that paper is WHOOP minus ECG.

WHOOP reports resting heart rate from sleep, which is a quiet, low-motion window. That is a different measurement from a seated clinic pulse.

Exercise Heart Rate

The WHOOP heart-rate trials cited above were recorded during sleep. Exercise is where wrist photoplethysmography meets motion.

Steady-state cardio (running, cycling, swimming): A repeating arm swing is the easier pattern for an optical sensor to separate from the pulse. Running tends to be steadier than cycling with a varying hand position.

High-intensity interval training (HIIT): Rapid rises and falls are harder because the processing pipeline lags the true rate. Short peaks can be missed.

Strength training and grip-intensive exercise: This is the weakest case for a wrist optical sensor, for WHOOP and for other wrist devices. Wrist flexion and a tight grip distort the signal. The technical reasons are in the PPG motion artifact removal guide.

Bicep strap: WHOOP's body strap moves the sensor to the bicep or another site. The brachial artery is larger than the vessels on the back of the wrist, and the upper arm has less tendon motion during many lifts. The validation studies on this page used the wrist.

Is WHOOP HRV accurate?

Yes for the overnight average, with a wider error than heart rate. The three peer-reviewed comparisons below all use RMSSD, or its natural log, against an ECG reference in healthy adults. They do not all use the same WHOOP generation or the same sleep window.

Dial et al. (2025): WHOOP 4.0 at home

Dial and colleagues published in Physiological Reports (doi:10.14814/phy2.70527). Thirteen healthy adults wore a Polar H10 and several consumer devices during home sleep, 536 nights in total across devices. WHOOP 4.0 HRV was available for 289 nights. All devices in the study report HRV as RMSSD. For WHOOP, Table 3 gives:

• Mean bias −0.78 ± 5.98 ms
• Limits of agreement −12.50 to 10.94 ms
• Mean absolute error 4.17 ± 4.33 ms
• MAPE 8.17 ± 10.49%
• Pearson r 0.96
• Lin's concordance 0.94

On the authors' scale, concordance from 0.90 to 0.95 is moderate. Their conclusion places WHOOP 4.0 HRV in the intermediate group, behind the two Oura rings in that study. The Polar H10 value was averaged over the whole night in Kubios. WHOOP's displayed HRV is weighted toward the last slow-wave sleep stage, so the comparison mixes sensor error with a different window. The authors report no competing interests. The Air Force Research Laboratory funded the study.

Miller et al. (2022): WHOOP 3.0 in the laboratory

Miller, Sargent, and Roach (Sensors, doi:10.3390/s22166317) recorded one night in 53 adults. WHOOP staff, kept unaware of the electrocardiogram, supplied R-R intervals for the whole sleep period. The researchers then computed RMSSD themselves, matched to the ECG window. Table 6, for WHOOP 3.0 HRV:

• Bias −4.5 ± 3.9 ms (WHOOP lower than ECG)
• Absolute bias 4.7 ± 3.6 ms
• Limits of agreement ±7.6 ms
• ICC 0.99

The footnote on that table marks WHOOP as raw manufacturer data. The authors' research group receives research support from WHOOP Inc. The paper states that WHOOP was not involved in the design, conduct, or reporting. This RMSSD is a whole-sleep calculation from supplied intervals. It is not automatically the number in the WHOOP app.

Bellenger et al. (2021): WHOOP 2.0 during slow-wave sleep

Bellenger et al. (Sensors, doi:10.3390/s21103571) studied 6 healthy adults (3 men, 3 women, age 22.9 ± 3.4 years) over 15 sleep opportunities with WHOOP 2.0. HRV was Ln RMSSD and RMSSD during slow-wave sleep. Bias is WHOOP minus ECG.

The abstract reports the lowest Ln RMSSD bias as 1.66 ± 1.80%, with limits of agreement ±5.93%, for a 200 ms (Kubios Strong) filter and for WHOOP's proprietary filter. In the results, proprietary-filter Ln RMSSD percent bias was ≤3.25 ± 1.53% across the two slow-wave definitions, with limits of agreement ≤6.59%. That bias approached or exceeded the smallest worthwhile change they calculate for Ln RMSSD (1.5–6.5%). For RMSSD during polysomnography-defined slow-wave sleep, the proprietary filter gave percent bias 12.74 ± 6.53% and limits of agreement 25.31%, which is an over-read relative to ECG. The paper's conclusion: acceptable agreement for heart rate, and HRV bias and limits that should be read against their own precision.

One co-author's position was sponsored by WHOOP. The paper states that the sponsorship began after data collection.

A nightly RMSSD is a personal baseline. Population ECG means by age, from a 5-minute supine recording, are on the good HRV by age chart. What HRV measures defines the metric. The HRV chart by age and the SDNN page cover related statistics.

How accurate is WHOOP ECG?

WHOOP MG can take an ECG. The accuracy figures below are from the FDA 510(k) summary of the sponsor's trial. A separate peer-reviewed accuracy paper for this feature is not among the sources on this page.

The FDA cleared the WHOOP ECG Feature (1.0) on 4 April 2025, 510(k) K243236, as Class II electrocardiograph software for over-the-counter use (21 CFR 870.2345, product code QDA). The substantial-equivalence table names the compatible strap as WHOOP MG. It is a software application that records, stores, and displays a single-channel ECG qualitatively similar to Lead I. The user records about 30 seconds on the wrist strap. On a classifiable waveform the feature labels atrial fibrillation, normal sinus rhythm, low heart rate (≤50 bpm), or high heart rate (≥100 bpm). The summary says it is for adults 22 and older, is not recommended for people with other known arrhythmias, and is informational. It says the user is not intended to take clinical action from the output without a qualified professional, and that the waveform does not replace diagnosis or treatment.

Clinical testing in the summary compared the feature with a cardiologist's reading of a simultaneous 12-lead ECG in about 540 subjects: 255 enrolled in the atrial-fibrillation cohort, 285 in the normal sinus-rhythm cohort, mean age 60.4 years. On classifiable recordings, sensitivity for atrial fibrillation was 96.2% and specificity for sinus rhythm was 99.4%, both for heart rates of 50–150 bpm. The feature called 11% of recordings inconclusive. When those inconclusive recordings are counted, the probability of an atrial-fibrillation result for a subject in atrial fibrillation was 87.47%, and the probability of a sinus-rhythm result for a subject in sinus rhythm was 96.59%. The summary says real-world performance for inconclusive and poor recordings may differ.

Age-group sensitivity in the summary ranged from 95.1% to 100%, and specificity from 97.6% to 100%. Specificity and sensitivity were 99.3% and 98.9% in women and 99.6% and 95.4% in men. For participants identifying as American Indian or Alaska Native the summary lists specificity and sensitivity of 100% and 71.4%, and attributes the lower estimate to the small sample. A visual comparison with the 12-lead reference judged waveform quality acceptable 99.4% of the time. The summary also says WHOOP compared PR interval, QRS duration, R-wave amplitude, and RR interval with the 12-lead ECG. It does not publish those differences, so this page does not state a millisecond or millivolt error for the trace.

The predicate device in the letter is the Apple ECG App 2.0. Clearance is a substantial-equivalence decision for that intended use. It is not a continuous rhythm monitor, and it is not the PPG heart-rate or HRV measurement described above.