Watch VO2 Max Estimates vs Lab Testing: How Far Off Are They
Your watch told you your VO2 max was 48. Three weeks later it says 44. You have not stopped training. Nothing has changed except the number.
Here is what is actually happening, and how far a wrist estimate sits from a measured result.
How a watch arrives at a VO2 max number
A watch does not measure oxygen. Measuring oxygen uptake requires collecting your expired air and analysing its composition, which requires a mask.
What a watch has is heart rate from an optical sensor, pace or power from GPS or a pod, and your entered age, sex and weight. It infers aerobic capacity from the relationship between external work rate and heart rate response. Move faster at a lower heart rate and the algorithm raises your estimate.
That inference is reasonable in principle. The problem is everything it has to assume.
Your maximum heart rate. Usually estimated from your age unless you have recorded a genuine maximum. If the assumption is out by 10 beats, the estimate is out with it.
Your resting heart rate. Measured, but it drifts with sleep, alcohol, illness and stress.
Mechanical efficiency. The algorithm assumes a standard oxygen cost for a given pace. Running economy varies substantially between individuals.
The quality of the input data. Wrist optical heart rate degrades during high intensity work, cold weather and irregular arm movement. GPS pace degrades under tree cover and in urban canyons.
Terrain and conditions. Heat, humidity, wind and gradient all shift the heart rate to pace relationship without changing your capacity at all.
Change any of those and the number moves. None of them are your aerobic capacity.
What validation studies find
Two recent studies put Apple Watch estimates against laboratory gas analysis in the same participants.
A 2025 validation study in Dublin tested 28 participants against indirect calorimetry. The watch underestimated VO2 max by a mean of 6.07 mL/kg/min. Mean absolute percentage error was 13.31%, and the Bland Altman limits of agreement ran from 6.11 mL/kg/min below to 18.26 mL/kg/min above the true value (Lambe et al., PLOS ONE).
A 2024 study of the Apple Watch Series 7 tested 19 participants across two separate sessions. Mean bias was 4.51 mL/kg/min below the laboratory value, overall mean absolute percentage error was 15.79%, and the intraclass correlation with laboratory testing was 0.47, which is poor agreement (Caserman et al., JMIR Biomedical Engineering).
The second study did something more useful than reporting an overall average. It split participants by fitness level.
Fitness group | Lab VO2 max | Watch VO2 max | Bias | Error rate |
Poor (n=3) | 35.1 | 38.9 | +3.8 | 10.7% |
Good (n=11) | 44.8 | 41.4 | -3.4 | 14.6% |
Excellent (n=5) | 54.7 | 42.7 | -12.0 | 21.5% |
Read the bottom row. In the fittest participants the watch was out by 12 mL/kg/min, and every one of the five was underestimated. The watch effectively compressed a 54.7 average down to 42.7.
This is the pattern that matters most for anyone who trains. The better your aerobic capacity, the worse the estimate gets, and it errs downwards. If you are fit and your watch says you are merely good, that may be the algorithm rather than you.
The trap in "validated"
You will see meta-analyses cited showing wearables are accurate. One pooled 14 studies across multiple brands and found a pooled bias of just 0.09 mL/kg/min, which sounds excellent.
The same analysis reported limits of agreement running from 9.92 below to 9.74 above the criterion value.
Those two figures are not in conflict. They are describing different things. A pooled bias near zero means that across a large group, overestimates and underestimates cancel out. The limits of agreement describe what happens to one individual. A device can be almost perfectly accurate on average and still be wrong by 10 mL/kg/min for you specifically.
Population accuracy is not individual accuracy. That distinction is the whole argument.
What laboratory testing gets you that a watch cannot
Measurement error in laboratory indirect calorimetry has been estimated at around 5%, with one meta-analysis of 39 studies reporting a mean standard error of 2.58 mL/kg/min. So a lab test is not perfect either. It is roughly three times more precise, and more importantly it produces things a wrist estimate structurally cannot.
Your actual maximum heart rate, measured at the point of volitional exhaustion rather than assumed from your age.
Both ventilatory thresholds, which is what training zones should be built from. A watch has no way to detect a ventilatory threshold because it cannot see your breathing.
Respiratory exchange ratio, which shows your fuel mix at every intensity and confirms whether you actually reached a true maximum.
Verification that the test was maximal. A watch estimate has no maximal criteria to meet. It produces a number regardless.
When a watch estimate is genuinely useful
This is not an argument that wearable estimates are worthless. They are useful for the thing they are good at.
Use it for: long term trend direction, spotting a sustained drop that might indicate overtraining or illness, and rough population placement if you have never been tested.
Do not use it for: setting training zones, tracking whether a 12 week block worked, comparing yourself against normative tables with any precision, or any decision where being out by 10 mL/kg/min would change what you do.
A practical approach is to measure properly once to get a real baseline and real thresholds, then let the watch track direction between tests. The lab number calibrates the wrist number. The wrist number gives you continuity.
Common questions
Why does my number jump after one bad run? Most algorithms update from qualifying activities, and a single session in heat or with poor sensor contact can shift the estimate. It is measuring your run conditions, not your physiology.
Is a chest strap better than the wrist sensor? For heart rate accuracy, yes, particularly at high intensity. It improves one input to the estimate. It does not make the estimate a measurement.
My Garmin and my Apple Watch disagree. Which is right? Neither is measuring anything. They use different proprietary algorithms with different assumptions, so disagreement is expected. The disagreement itself tells you how much uncertainty is in both.
Should I stop looking at it? No. Just stop treating it as a result. Treat it as a trend line with a wide confidence band.
The short version
Wrist-based VO2 max estimates carry published mean absolute percentage errors of roughly 13 to 16%, and the error grows in fitter people, reaching 12 mL/kg/min of underestimation in the excellent fitness group of one validation study. The number moves because its inputs move, not because your capacity does. Use the watch for direction and a measured test for decisions.
If you want to know where a real number sits, see VO2 max by age and sex.
A measured VO2 Max test at Precision Body Lab is $349, or $449 with a DEXA scan. Book online or call 1300 910 163.
Related reading
Sources
Lambe R, O'Grady B, Baldwin M, Doherty C, et al. Validity of Apple Watch VO2 max estimation against indirect calorimetry. PLOS ONE, 2025. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0323741
Caserman P, Yum S, Göbel S, Reif A, Matura S. Assessing the Accuracy of Smartwatch-Based Estimation of Maximum Oxygen Uptake Using the Apple Watch Series 7: Validation Study. JMIR Biomedical Engineering, 2024. https://pmc.ncbi.nlm.nih.gov/articles/PMC11325102/
Meixner B, Filipas L, Holmberg HC, Sperlich B. Zone 2 Intensity: A Critical Comparison of Individual Variability in Different Submaximal Exercise Intensity Boundaries. Translational Sports Medicine, 2025. https://pmc.ncbi.nlm.nih.gov/articles/PMC11986187/
Comments