How Accurate Is VO2 Max on Apple Watch and Garmin? What the Validation Studies Show

Short answer: wearable VO2 max estimates usually land within about 5% to 15% of a lab test, and the direction of the error depends on who is wearing the watch.

In lab studies, Apple Watch read about 6 mL/kg/min low in healthy adults, Garmin read about 5 low in endurance athletes, and a consumer smartwatch read 14.5 high in less fit, mostly middle-aged adults [1][2][3][4]. That is good enough to track whether someone’s fitness is improving. It is not good enough to tell one person precisely where they stand.

The question matters more each year. VO2 max is one of the strongest predictors of how long people live [5][6], Apple now uses it in its Health Age feature [7], and insurers and longevity apps increasingly treat it as a headline number. This post covers how watches estimate it, what the validation studies show, why the error flips with fitness, and how a product should use it.

Why does VO2 max matter?

VO2 max is the maximum rate at which the body can take in and use oxygen, the standard measure of cardiorespiratory fitness. It is measured in a lab by analyzing breath during exercise to exhaustion.

The case for caring about it is mortality. In 122,007 patients who had treadmill tests at the Cleveland Clinic, adjusted all-cause mortality fell steadily with fitness, and the lowest fitness group had about five times the risk of the elite group [5]. A meta-analysis of 33 cohort studies found that each 1 MET higher fitness level, about 3.5 mL/kg/min of VO2 max, was associated with 13% lower all-cause mortality and 15% lower cardiovascular events [6].

That is why fitness is moving into consumer health products. Apple’s Health Age, announced in September 2026, uses VO2 max alongside resting heart rate, sleep and HRV [7]; the rest of that announcement is covered in Apple’s September health features. Apple has shown Cardio Fitness levels and low-fitness notifications since watchOS 7 [8]. Oura’s API exposes a VO2 max estimate and a cardiovascular age [9]. The longevity app market is built on metrics like this one.

How do watches estimate VO2 max?

Nobody runs to exhaustion wearing a gas mask in daily life, so watches estimate VO2 max from submaximal exercise.

Apple Watch generates an estimate after outdoor walks, runs or hikes on relatively flat ground, under 5% incline, with adequate GPS and heart rate signal and a heart rate rise of about 30% above resting [10]. It models how heart rate responds to pace, using the wearer’s age, sex, weight and height.

Garmin uses Firstbeat’s method, which analyzes the relationship between heart rate and speed at multiple points during a run, discards unreliable segments, and updates the estimate after each qualifying session [11].

Both rely on the same principle: a fitter person covers a given pace at a lower heart rate. Both also rely on assumptions, most importantly about each person’s maximum heart rate and how efficiently they move. When those assumptions are wrong for an individual, the estimate is wrong in a predictable direction.

What do the validation studies show?

Apple Watch. In a 2025 study, 30 adults wore an Apple Watch for 5 to 10 days to generate an estimate and then did a lab treadmill test. The watch underestimated by 6.07 mL/kg/min on average, with a mean absolute percentage error of 13.3% and wide limits of agreement [1]. A 2026 follow-up with Series 10 in 40 healthy adults found almost the same: 6.25 low, 13.2% error. It placed only 14% of participants in the correct fitness percentile band, and another 34% one band away. The authors concluded the watch showed poor individual agreement but could be useful for population-level research [2].

Garmin. Firstbeat’s own validation, on 2,690 runs from 79 marathon runners tested four times over six to nine months, reports about 5% error [11]. An independent 2025 study of 35 endurance athletes using a Forerunner 245 found 7% to 8% error, reading 4.7 mL/kg/min low overall. In moderately trained athletes the error was about 3% to 4%; in the most highly trained it rose to about 10%, reading 6.3 low [3]. A study of the fēnix 6 in a general population of 19 people found about 7% error [12].

The systematic view. A 2025 review of 13 studies, most on Garmin devices with Firstbeat’s algorithm and many with chest-strap heart rate, found estimates valid or acceptable in seven, across untrained adults, recreational athletes and team-sport professionals, with elite endurance athletes the weakest group [13].

Less fit people. The most important recent study is the least flattering. In 203 mostly middle-aged adults, median age 55, with a median lab VO2 max of 22 mL/kg/min, a consumer smartwatch worn for three months overestimated by 14.5 mL/kg/min [4]. Its estimates correlated only moderately with the lab result (r = 0.65) but very strongly with a formula using just age, sex, weight and height (r = 0.92). The error was largest in people whose fitness was lower than those demographics predicted. The study was a preprint when published, and the device was a single consumer watch, but the pattern is a warning for any product aimed at a general population.

Is wearable VO2 max still useful?

Yes, for the right questions.

• Change over time, same person, same watch. A consistent bias largely cancels out when you compare someone with themselves. A rising estimate over three months of training is more trustworthy than any single value. The same logic applies to validating any health score.
• Populations and research. Averaged over many people, individual errors matter less, which is why the Series 10 authors saw a role in population studies [2].
• Engagement. A fitness number that moves with effort is motivating, and the mortality evidence gives it meaning.

It is weak for:

• Placing one person in a fitness percentile. One in seven in the correct band is not a classification tool [2].
• Pricing, underwriting or clinical decisions on a single reading.
• Sedentary people who rarely walk briskly outdoors, because the watch has fewer qualifying sessions and leans more on demographics [4][10].

How should a product use wearable VO2 max?

• Keep the method with the value. HealthKit’s VO2 max samples carry a test type: maximal exercise, submaximal prediction, non-exercise prediction or step test [14]. Health Connect’s VO2 max record carries a measurement method such as metabolic cart, heart rate ratio, Cooper test or Rockport test [15]. A lab result and a watch estimate should never share one series without that label.
• Show a band or a trend, not a precise number. Present the estimate as a range or a fitness band with a note that it is an estimate, and emphasize change over weeks.
• Watch for the unfit-but-average case. If an estimate looks close to what age and sex alone would predict, treat it as low information, especially for users with few outdoor workouts.
• Do not mix vendors. Apple, Garmin, Oura and others use different methods. A user who switches devices will see a step change that is not a change in fitness. The same applies to normalizing data across providers.
• Encourage the input the estimate needs. Regular brisk outdoor walks improve both the estimate and the fitness it measures.

The short version

VO2 max predicts mortality strongly [5][6], and watches now estimate it for millions of people. In lab studies, Apple Watch read about 6 mL/kg/min low with 13% error and placed only 14% of people in the right fitness band [1][2]; Garmin’s Firstbeat method read about 5 low in endurance athletes with 7% to 8% error [3]; and in less fit adults a smartwatch read 14.5 high, barely improving on a demographic formula [4]. Use it for trends in the same person, label it as an estimate with its method, and do not classify, price or treat anyone on a single value.