How Accurate Is Garmin HRV? Fenix 6 Nocturnal Study

Short answer: In one home-sleep study, a Garmin Fenix 6's nocturnal RMSSD lined up with a Polar H10 chest strap at a concordance of 0.87, with a mean bias of −1.84 ms and limits of agreement from −15.218 to 11.60 ms, on 150 nights. The same paper's abstract calls that agreement poor. Newer Garmin beat-to-beat files are a different product generation, and the public checks on those files are a one-night manufacturer example and a Labfront commentary that does not report a sample size.

This is a reading of those sources. It is not a buying guide, a medical clearance, or advice about training.

The study to quote first

Dial, Hollander, Vatne, Emerson, Edwards, and Hagen published the comparison in Physiological Reports in 2025 (13(16), doi:10.14814/phy2.70527, PMC12367097). Thirteen healthy adults, six of them women, wore a reference strap and several wearables at once during sleep at home. Anyone with at least 10 nights was included. The combined record was 536 nights.

The reference was a Polar H10, single-lead ECG at 1000 Hz. R-R intervals were processed in Kubios. Every wearable in the HRV table reports HRV as RMSSD.

The Fenix 6 uses Garmin's Elevate V3 optical sensor. Garmin does not publish the raw PPG sampling rate; the paper quotes the manufacturer that the rate "varies, and may depend on the level of activity of the user." Heart rate during activity is a processed value each second. During sleep the watch measures heart rate intermittently. Resting heart rate, in Garmin's definition as cited by Dial, is the lowest 30-minute average in 24 hours. Because that window has no timestamp they could line up with the strap, Garmin is absent from the resting-heart-rate tables.

HRV on the Fenix 6 is a different path. The paper says the watch measures it continuously through sleep, in 5-minute windows, and averages those windows across the sleep the watch detected, using Firstbeat / Garmin processing.

The methods do not say whether Bland-Altman bias is wearable minus Polar H10 or Polar H10 minus wearable. The table below copies the printed numbers and does not assign a sign convention the paper left unnamed.

Table 3, nocturnal RMSSD

Dial et al., Table 3, "Heart rate variability accuracy." Values are printed as mean ± SD where the paper uses that form. Limits of agreement are the pair in the table. Nights differ by device because people did not wear every device every night.

The abstract ranks HRV agreement as highest for the Oura rings, moderate for WHOOP 4.0 (CCC 0.94, MAPE 8.17 ± 10.49%), and poor for both Garmin (CCC 0.87, MAPE 10.52 ± 8.63%) and Polar (CCC 0.82, MAPE 16.32 ± 24.39%). In the results text, Garmin's MAPE was higher than Oura Generation 3 (p = 0.0256), Oura Generation 4 (p = 0.0133), and WHOOP 4.0 (p < 0.0001). A later z-score table in the same paper prints a Garmin concordance of 0.77. That is a normalized re-analysis, not a replacement for Table 3.

Table 3 is one mean bias per device. It is not a slope of error against the size of RMSSD. A magnitude-dependent pattern is discussed below only as Labfront's reading of the figure.

What this row does not cover

• Device generation. The optical watch in the table is a Fenix 6 with Elevate V3. A later Garmin is a different instrument.
• Raw sampling. The PPG rate during sleep is unpublished in this paper.
• Window. The exported HRV is an average of 5-minute sleep windows, which is a different object from a single 5-minute supine lab recording. The RMSSD guide separates those.
• Reference class. The strap is a single-lead ECG, processed in Kubios, in healthy adults at home. Thirteen people, even across 150 Garmin nights, is a narrow sample.
• Sign of the bias. Unknown from the methods text.
• Daytime movement. This table is nocturnal. Agreement between pulse rate variability and ECG HRV is a separate literature, and it is weaker once people move.
• Other brands' rows above are context from the same table. They are not a ranking this site computed.

Other published HRV rows, across brands, sit on the HRV accuracy leaderboard.

Garmin's own one-night example

Garmin Health's Enhanced BBI white paper (November 2023, PDF) is a manufacturer document. The worked night is one subject: a Venu 2 Plus on the left wrist and a Firstbeat Bodyguard 2 ECG. Age 56, male, BMI 25.9. The analyzed sleep ran from 10:06:44 pm to 5:56:00 am. The file had 26,248 beats, of which 24,468 (93.13%) were high confidence.

On those high-confidence beats, mean R-R interval was 1078.6 ms and mean beat-to-beat interval was 1078.0 ms (standard deviations 92.90 ms and 93.99 ms). Beat error was nearly zero mean, 0.506 ms, with a standard deviation of 8.55 ms. Garmin's text says 68.2% of beat errors were, on average, under 0.79% of the beat-to-beat magnitude, which is their reading of that 1 SD. Pearson's r between high-confidence beat-to-beat intervals and R-R intervals was 0.975. Poincaré SD1 was 124.38 ms (R-R) and 125.88 ms (beat-to-beat); SD2 was 42.32 ms and 42.65 ms. Confidence in the file is a 0 or a 1. The paper says the devices are not medical devices.

That is one night. It is not a multi-subject validation, and the word "offset" does not appear as a summary RMSSD bias in the PDF text we checked.

Labfront's Enhanced BBI commentary

Labfront's article on Enhanced BBI (labfront.com) describes the company's own month-long overnight collection: a Garmin Venu 3 worn with a Movesense chest ECG. They computed average 5-minute nocturnal RMSSD. For legacy beat-to-beat intervals and for Garmin Connect, they say they reproduced a pattern they attribute to the 2025 study: overestimation when RMSSD was low and underestimation when it was high. Dial's Table 3, quoted above, is a single mean bias, so that slope is Labfront's figure commentary.

For Enhanced BBI, after low-confidence beats were removed and each 5-minute window was required to be 70–90% high confidence, they describe the Bland-Altman cloud as flat, with a mean difference of about +8 ms. They suggest the remaining offset is physiological pulse-rate variability rather than an algorithm error. The article does not state a participant count or a night count. The +8 ms figure stays Labfront's, and it is not a peer-reviewed result.

Pulse versus ECG intervals is the place to read why a pulse series can sit slightly off an electrical series even when beat detection is doing its job. Sleep regularity is a different question: it scores the clock, not the beat-to-beat gaps inside one night.