How accurate is Apple Watch Cardio Fitness, and Garmin’s VO₂max?

A watch does not measure oxygen. It estimates VO₂max from exercise data. Apple’s estimate, for example, needs an outdoor walk, run or hike on a grade under 5%, and reports values between 14 and 60 mL/kg/min.

A 2022 meta-analysis of 14 validation studies split the algorithms in two. Those that used exercise data were unbiased on average: a mean bias of −0.09 mL/kg/min. Those that used resting data significantly overestimated VO₂max.

Unbiased on average is not the same as accurate for you. The limits of agreement, the range that about 95% of individual differences fall within, ran from −9.92 to 9.74 mL/kg/min. The authors call the estimates good for populations and poor for individuals.

Apple’s white paper tested the Series 4 on a design set of 534 people and a validation set of 221, with a mean age of 55 and 38% over 65. The average error was 1.2 ± 4.4 mL/kg/min in the design set and 1.4 ± 4.7 in the validation set. Test-retest reliability, the agreement between repeated readings, had an ICC of 0.86 (0.80–0.90).

The catch is the reference. Apple did not compare the watch with a measured maximum. It compared it with a projection: heart rate and oxygen use at submaximal effort, extended in a straight line to an age-predicted maximum heart rate. That reference is itself an estimate, and the paper was written by the vendor.

Three small studies have tested Apple Watch against VO₂max measured by gas analysis, and all three found it read low:

• 2025, treadmill, 30 adults. The watch read 6.07 mL/kg/min low on average (95% CI 3.77–8.38). Individual differences ran from about 18 too low to about 6 too high. Mean absolute percentage error 13.31%.
• 2026, Series 10, 40 adults. 6.25 low on average, with a similar individual range. Only 5 people (14%) landed in their correct FRIEND percentile band; another 12 (34%) were one band off.
• 2024, cycle, Series 7, 19 adults. Lab 45.88 against watch 41.37 mL/kg/min. ICC 0.47, error 15.79%. The watch overestimated low-fit people and underestimated high-fit ones.

These are small samples, and the test mode differs: cycle tests read lower than treadmill tests. A 2026 living review found only one Apple Watch VO₂max study it could include, so there is no pooled estimate yet.

The gap between Apple’s +1.4 and the independent −6 is not explained by any of these papers. The different references, a projection against a measured maximum, are the obvious place to look.

In a 2025 study of 19 adults wearing a Garmin Fēnix 6, the mean absolute percentage error was 7.05% and Lin’s concordance was 0.73. That met the authors’ own criteria for acceptable accuracy: concordance above 0.7 and error under 10%.

An earlier study of five wrist devices reported errors above 10% for all of them except a Garmin Forerunner 920XT for VO₂max. We have not checked that figure against the paper.

Fewer than 20 people, and results that apply to one device and one firmware version, are not enough to call Garmin accurate. None of the studies here compares a current Garmin with a current Apple Watch.

Garmin also places your number in fitness categories. Those come from the Cooper Institute table, which is predicted from treadmill time and reads higher at older ages than the measured FRIEND norms Apple uses.

An error of 10 mL/kg/min sounds modest until you set it against the norms. In the FRIEND table, men aged 50–59 run from 27.1 at the 25th percentile to 32.6 at the median and 39.7 at the 75th. Ten points either way can carry a man from below the median to above the 75th percentile.

For women aged 60–69 the whole table is narrower: 14.6 at the 10th percentile, 20.0 at the median, 27.0 at the 90th. There, a 10-point error covers most of the range.

A watch reading repeated over months may still show direction. Apple’s own test-retest figure suggests readings are consistent. But the studies above tested single readings against the lab; none tested whether a watch tracks real change accurately.

Longstanding can read VO₂max from Apple Health or Health Connect, and it says which source each value came from. It scores that value against the Cooper Institute table, the one Garmin uses, not against FRIEND, the one Apple uses. The same watch value can read as low fitness in one place and fine in the other.

If your VO₂max comes from a watch, the app’s percentile can be a band or two off. A field test or a lab result, repeated the same way every quarter, is the better input. The 12-minute run and the one-mile walk are the two field options, and both are on the VO₂max test page.

A low watch reading is not a diagnosis, and none of this is medical advice.