Is Garmin's VO2 Max Accurate? 42 Athletes, 14,919 Steps

Is Garmin's VO2 max accurate?

We don't know, and neither does anyone else quoting a number at you. Not one athlete in our data has a metabolic-cart test to compare against, so we cannot check the watch's absolute accuracy, and we are not going to pretend otherwise.

What we can check is how the number behaves. We pulled every VO2 max value our platform holds for athletes with at least 180 days of history and watched the series move. That turns out to be the more useful question anyway, because the way this metric behaves tells you exactly how much weight it can carry.

The honest problem with the question

VO2 max is a lab value. Measuring it properly means a mask, a metabolic cart, a treadmill, and someone raising the grade until you cannot continue. Your watch does none of that. It infers the number from the relationship between your GPS pace and your heart rate during steady outdoor efforts, using an algorithm from Firstbeat Analytics.

Firstbeat's own white paper claimed roughly 5% mean absolute percentage error across 2,690 runs from 79 lab-tested runners. Worth noting: that document is no longer publicly hosted. Firstbeat was acquired by Garmin in 2020 and the URL the figure was cited from for years now fails to resolve. The number is still repeated across the internet, largely by people who can no longer read the source.

Independent work is more useful anyway. A 2025 systematic review in Frontiers in Sports and Active Living pooled 13 studies covering 273 participants that compared wearable estimates (mostly Garmin, mostly running the Firstbeat algorithm) against laboratory graded exercise tests. Seven of those studies judged wearables acceptable for estimating VO2 max. On error: one study came in under 5%, four landed between 5% and 10%, and one reached 15.78%.

That range is the honest headline. For a runner sitting at 50, a 5% error is 2.5 points and a 10% error is 5 points — enough to move you a full fitness category. The same review found the picture worse for lactate threshold, where only one study has demonstrated validity at all.

Here is a correction we owe readers. Two earlier gneta articles, why your Garmin VO2 max is wrong and how to improve your Garmin VO2 max, told you the estimate was "typically within 3 to 5% of lab-tested values" without saying where that came from. It came from the manufacturer's own paper, and it sits at the optimistic edge of what independent research reports. We should have cited it, and we should have said that a vendor validating its own algorithm is not the same as an independent test on you. Both articles now carry the range above instead.

What we measured: 42 athletes, 180+ days each

Our platform stores the daily VO2 max value Garmin reports for every connected athlete. For study S3, computed on 1 August 2026, we took everyone with at least 180 days of history: 42 athletes. Six further accounts were excluded because they were created during a May 2026 incident window and carried near-identical multi-year histories under distinct Garmin accounts, which makes them non-independent rather than useful.

The usual caveat, and it matters most on a page about accuracy: these are self-selected Garmin owners who connected a training-analytics app. They are probably fitter and more data-curious than the average watch wearer. The cohort's median current VO2 max is 51.9 (p25 47.7, p75 55.5), which is a fairly fit group by any population standard.

One more limitation before the findings. This is an observed-day series. When an athlete goes several days without syncing, a single observed step can bundle several underlying updates, so our measured step sizes are an upper bound. If we are wrong, the real algorithm moves in even smaller increments, even more often.

Finding 1: it moves constantly, in whispers. Across the cohort, the value changed a median of 16.0 times per 90 days — roughly every sixth day. The middle half of athletes sat between 10 and 25 changes per 90 days. When it does move, the median step is 0.2 points (p25 0.1, p75 0.5, p90 1.0), measured across 14,919 observed steps. Nine steps in ten are one point or smaller.

Finding 2: over a year, it moves for real. Comparing each athlete's value to their own value twelve months earlier, the median absolute change was 2.0 points (n = 29). Only 31% of athletes stayed within ±1 point of where they were a year ago, and 51.7% stayed within ±2. Roughly half the cohort moved more than two points in a year.

Finding 3: the full range is wide. Over each athlete's complete multi-year history, the spread between their lowest and highest recorded value had a median of 9.9 points (p25 7.7, p75 13.0). That is not a stable physiological constant being read out. That is a tracking estimate with years of room to wander.

Put those together and the picture is clear. The number is far too twitchy to read day to day, and far too mobile across a year to dismiss as noise. It behaves like a slow-moving trend line, because that is what it is.

Why 0.2 is a number you should never look at

This is where I will state an opinion. Showing a 0.2-point change to a user is a product decision, and I think it is a bad one. No method of measuring VO2 max, in a lab or on a wrist, resolves anything close to 0.2 ml/kg/min. A step that small is the algorithm's internal smoothing surfacing itself, and the only thing it reliably produces in the wearer is anxiety.

Every forum thread that begins "my VO2 max dropped by 1 and I don't know why" is a response to a display choice. A 1-point step sits at the 90th percentile of all moves in our data, which sounds notable until you remember it is still smaller than the year-over-year change half this cohort experienced anyway. Show one value per month and the wearer loses nothing.

What the race-predictor data adds

Race predictions are computed downstream of your VO2 max estimate, so they inherit its errors. In our race predictor accuracy study, 11 real races from 6 athletes came in at a median miss of +0.6% against the prediction from the day before.

That is weak but real evidence that the VO2 max input is not wildly wrong, at least for this cohort. If the underlying fitness estimate were off by 10%, it is hard to see how the downstream race times would land within a percent of reality on race day. Eleven races is a case series, and the running-economy layer between VO2 max and predicted pace could be absorbing some of the error, so treat it as a hint rather than a validation. You can cross-check the same logic yourself by putting a recent result into our race pace predictor and comparing it against what your watch says.

What the estimate cannot see

Because the number is derived from pace against heart rate, anything that changes one without changing the other corrupts it:

• A wrong max heart rate. This is the big one, and Firstbeat quantifies it: a max HR set 15 bpm too low produces roughly 9% error in the VO2 max result, and 15 bpm too high roughly 7%. Fix this before anything else, using our heart rate zones setup guide.
• Heat and humidity. Cardiac drift raises heart rate at the same pace, so a hot month reads as a fitness decline.
• Hills. Uphills inflate heart rate relative to pace; downhills do the reverse. The algorithm corrects for grade, imperfectly.
• Wind and drafting. A tailwind or a group run gives you free pace the watch attributes to your aerobic system.
• Optical HR errors. A wrist sensor reading low during hard efforts hands the algorithm a flattering pace-to-HR ratio.

None of these are bugs. They are the consequence of inferring a laboratory quantity from two signals that weather and terrain also move. A properly calibrated lactate threshold helps the model anchor your effort scale, but it does not make wind stop existing.

What it takes to move it, and why it stalls

The honest answer from our numbers: the unit of change is the training block, not the run. A median year moves this metric 2.0 points. If you are waiting for a session to shift it, you are watching the wrong timescale.

Stalls usually have one of four causes. The algorithm never gets a clean segment, because your runs are intervals and easy jogs with no steady 10–15 minutes in between. Your routes are hilly enough that the grade correction eats the signal. You are training through a hot stretch. Or you have genuinely plateaued, which happens to everyone eventually and is covered properly in our guide to improving your Garmin VO2 max.

If you want an independent read rather than a wrist estimate, run a Cooper 12-minute or 1.5-mile test and put the result through our VO2 max estimator. It is still an estimate, from a different set of assumptions, which is exactly what makes the comparison worth doing. Our VO2 max metric reference covers what the number represents physiologically.

When to ignore it entirely

Ignore any single-day change, always. Ignore the number during a heat wave, during the first two weeks on a new watch or sensor, during illness, and during a block where you have switched to noticeably hillier terrain. In all of those cases you are reading a measurement artifact.

Ignore it too when it disagrees with things that measure your actual training rather than your estimated physiology. If Training Status says productive and your sessions are showing the training effect you intended, a two-week VO2 max dip is not information. It is the algorithm catching up.

---

Frequently Asked Questions

Is Garmin's VO2 max accurate compared to a lab test?

We cannot answer this from our own data, because no athlete in our 42-person cohort has a lab test on file. The manufacturer's own validation reports around 5% mean absolute error against lab testing in a group of 79 runners. That is the best number available, it comes from the vendor, and it was measured on runners under study conditions rather than on you in a Tuesday headwind.

How much should my Garmin VO2 max change in a year?

In our cohort the median year-over-year change was 2.0 points in absolute terms. Only 31% of athletes were within ±1 point of their value a year earlier, and 51.7% were within ±2. A two-point move in either direction over twelve months is completely ordinary.

Why does my Garmin VO2 max only change by 0.2?

Because that is the algorithm's typical step. Across 14,919 observed changes in our data the median step was 0.2 points, with 90% of steps at 1.0 point or smaller. The value changed roughly every sixth day. Small, frequent adjustments are normal behavior, not a sign that your watch is stuck.

Does a 1-point VO2 max drop mean I lost fitness?

Almost certainly not on its own. A 1-point step sits at the 90th percentile of all moves we observed, so it is a large single step by the algorithm's standards while remaining small against the 9.9-point median range athletes cover over their full history. Check whether it was hot, hilly, or a race week before concluding anything.

Should I trust my race predictions if I do not trust my VO2 max?

They share a root, so a bad VO2 max produces bad predictions. But our race-day sample landed at a median 0.6% miss, which suggests the chain holds up better on a tapered start line than it does on any given training run.

---

The thing worth changing is not the algorithm. It is your sampling rate. Write your VO2 max down on the first of every month and look at twelve numbers a year, and the metric becomes what it always was underneath: a slow, noisy, directionally useful index of aerobic fitness. Check it after every run and you are reading 0.2-point rounding decisions as biology.

We would like to close the gap this article opens. If you have a genuine lab-measured VO2 max and a Garmin history to go with it, we want the pair. That comparison is the study nobody has run at scale, and it is the only thing that would let us answer the question in the title properly.