Why wrist heart rate fails in hard sessions

Wrist sensors do their best work at a steady effort. Ask for a quick change, or squeeze a barbell, and the error grows. The studies agree on that much. They don't tell you how your current watch behaves, because hardly anyone has tested it.

I don't test devices and I haven't worn any product named here. Each result names its authors, and I checked all of this as of 3 November 2026.

A tracker isn't a medical device. Wearable readings are estimates for general fitness and wellness, and they don't diagnose or treat any condition. If you have symptoms or concerns, talk to a clinician. If you have a heart condition, follow your care team's guidance on training zones rather than a watch reading. Etiwy and colleagues (peer-reviewed, 2019) found optical monitors less accurate than an electrode strap in 80 cardiac rehab patients.

Two things go wrong at once

The first is movement. Polar's own manual for the Vantage V3 (manufacturer guidance, not a test) says the sensor needs to stay still, and warns about "pressure or movement in muscles or tendons near the sensor". Gripping a bar does exactly that.

The second is lag. Fellrnr, an independent, non-peer-reviewed site, compares optical monitors against a chest strap across more than 1 million data points on old hardware such as the Garmin 225 and 235. It reports about 78% of readings within 3 bpm. It also reports that changes in intensity cause problems, sometimes with the sensor not noticing the change at all. One author, old watches, no date on the page. Still, the description fits the studies below.

Intervals: the weak spot

Two peer-reviewed studies come closest. Neither tests a current watch in a real running interval session.

Reddy 2018 (peer-reviewed, JMIR mHealth). 20 healthy adults, chest strap as reference. High-intensity bike intervals produced the worst errors: -11.4% (SD 35.7) for the Fitbit Charge 2 and -14.3% (SD 20.5) for the Garmin vívosmart HR+. Treadmill intervals were far closer, at -1.7% (SD 11.5) and -0.5% (SD 9.4). Both devices are obsolete.

Van Oost 2025 (peer-reviewed, Sensors). 24 students aged 18 to 27, skin types 1 to 3, against a 12-lead ECG. All wrist devices got worse during transitions, and sitting-to-walking often passed 8 to 12% error. The table is the full protocol at 10-second data.

Look at the missing-data column before you read the Garmin as the winner. It scored on about half the data. That protocol was also walking only, so it never reached interval intensity.

Here's the gap, in plain words. I found no large independent study of current flagship watches in structured running intervals, so I have no figure to give you for any current model in hard intervals.

The gym

Pooled numbers first. Zhang and colleagues (2020, peer-reviewed, 44 articles and 15 brands; abstract read, funding unchecked) put the mean difference from reference at -7.26 bpm in resistance training, with a confidence interval of -10.46 to -4.07. On a treadmill it was -0.51 bpm. The gym gap grew by 3 bpm for every 10 bpm rise in heart rate, so the harder you lift, the further off it reads. These averages also hide individual errors, because highs and lows cancel out.

Two newer lab studies:

• Lee 2026 (peer-reviewed, Sensors; funded by SOLUM Health Care, which employs one author): Apple Watch Series 7, Galaxy Watch 4, Fitbit Charge 5 and Garmin Vivosmart 5, on 62 healthy men with a mean age of 26.55, against ECG. Only the Apple Watch showed no significant difference from ECG in resistance exercise. Limits of agreement were about ±10 bpm. That's men only, and ±10 bpm is a wide band if you train by zone.
• Hajj-Boutros 2023 (peer-reviewed, abstract read, funding unchecked): 60 young adults with a Polar H10 as reference, across 5 activities including resistance. Variation was under 5% for the Apple Watch 6, 2.44 to 8.80% for the Polar Vantage V and 4.14 to 10.76% for the Fitbit Sense.

So treat gym readings as a rough guide. None of this carries over to a different model or firmware.

Cycling and trails

Zhang's pooled cycling figure is -4.55 bpm (95% CI -7.24 to -1.87). Bike intervals were the worst case in Reddy, as above.

On a hilly trail, Navalta and colleagues (2020, peer-reviewed, PLoS One) tested 21 healthy adults over a 3.22 km (2 mi) self-paced run, with a Polar H7 as reference. The Suunto Spartan with a chest strap had 2% error, and the Garmin Fenix 5 wrist sensor had 13%. Small study, one run, old hardware, and the reference is itself a strap.

Gym machines with moving handles look hard too. In Gillinov's 2017 study of 50 healthy adults (peer-reviewed, abstract read, funding unchecked), no optical device reached rc 0.80 on an elliptical with arm levers. The Polar H7 strap reached 0.996 overall. Those models are about a decade old.

What I'd do about it

For steady runs, try better fit and placement before you buy anything. Polar's manual suggests sliding the watch further up your arm and tightening the band for training.

For intervals, lifting and hard bike sessions, wear a chest strap if the number matters to you. Polar recommends one too, in that same manual, which is a little bit of a confession from a company selling wrist sensors. It does also sell straps.

An upper-arm sensor is the middle option. Schweizer and colleagues (2025, peer-reviewed, JMIR Cardio; no conflicts declared) tested 16 adults from lying still to HIIT against a Polar H10. A Polar Verity Sense on the upper arm had 1.43 bpm mean absolute error, and a Polar Vantage V2 on the wrist had 6.41 bpm. Every device there was a Polar and only 16 people took part, so it's a pointer and no more.

If you go that way, see the heart-rate monitor list for HIIT and strength and how to connect a strap to your watch. The gym and cycling picks start from the same problem.