Wrist or Chest Strap? How Accurate Optical Heart Rate Really Is
Wrist sensors track heart rate well on a steady run but drift on the bike, in sprints and during short intervals. What validation studies show, how a chest strap compares, and how to check your own device.
Sprint Summary
The short version — read this if you're short on time.
Most watches now measure heart rate with an optical sensor on the back of the wrist, while a chest strap reads the heart's electrical signal directly. Validation studies tell a consistent story. A 2020 meta-analysis of 44 studies found wrist devices were, on average, within about half a beat a minute of the reference during treadmill walking and running, but read about 4.6 beats a minute low on the bike and further out in resistance exercise. In a 2017 study of 50 adults, the chest strap agreed with an ECG almost perfectly, while optical devices ranged from good to poor, and only three of five were acceptable on the bike.
For a steady run or an easy ride, a modern wrist sensor is usually close enough. For short intervals, sprints, hard cycling, and any session where you set zones or pace from heart rate, a chest strap is the more reliable choice. These studies used devices that are now several years old, so the best way to know is to check your own watch against a strap on your own sessions. And a wearable isn't a medical device: chest pain, palpitations or a rhythm alert need a doctor.
A wrist sensor or chest strap is a training tool, not a medical device. Palpitations, chest pain or pressure, unexplained breathlessness, dizziness or fainting during exercise need a doctor, not a re-strapped watch or a firmware update.
An irregular-rhythm notification, or a heart rate that stays unusually high or low at rest with symptoms, is a reason to see your GP. In one study, wrist devices badly underestimated heart rate in people with atrial fibrillation, so a normal-looking number doesn't rule a rhythm problem out.
Medicines such as beta-blockers lower heart rate during exercise, so standard zone formulas and device-estimated zones may not fit. If you take heart-rate-affecting medication, agree your training intensities with your doctor.
If you have a pacemaker, implanted defibrillator or a known heart condition, ask your cardiology team which monitors and training intensities are suitable before relying on any wearable.
Full Distance
The complete research and analysis.
You head out for an easy run and your watch says 172 beats a minute. You can talk easily, so it isn't true, and ten minutes later it drops to 140. On the bike the next day, the trace looks smooth but strangely low during a set of hard efforts. Is the watch broken, is your heart doing something odd, or is this just what wrist heart rate does?
Most triathletes now get heart rate from an optical sensor on the back of a watch. TriForward's "Finding Zone 2: Heart Rate, Lactate, Talk Test or Breathing Threshold?" covers how to set zones. This guide is about the device: how accurate wrist heart rate is compared with a chest strap, where it goes wrong, and how to check your own.
One caveat first: device studies date quickly. The studies below tested devices from roughly 2015 to 2019. They show where optical heart rate tends to struggle, not how today's watch performs, so this guide names each study's year and device type rather than ranking models.
Two ways to measure a heartbeat
The chest strap: reading the electrical signal
A chest strap's electrodes pick up the electrical signal of each heartbeat, the same signal an ECG records, so it can time each beat precisely. That's why straps are used for heart rate variability, covered in TriForward's "Reading the Trend, Not the Number".
The optical sensor: reading blood flow
An optical sensor, technically called photoplethysmography or PPG, shines green (and sometimes red or infrared) light into the skin and measures how much is reflected back. Each heartbeat pushes a pulse of blood through the small vessels under the sensor, which changes how much light is absorbed. Software turns that rhythmic change into a heart rate.
The difficulty is that blood flow is not the only thing that changes the light signal. A 2020 study by Bent and colleagues in npj Digital Medicine described three main sources of error in optical sensors:
- Motion artefacts: movement of the watch on the wrist, and of the tissue under it, changes the signal.
- Signal crossover: the sensor can lock on to a repetitive movement, such as your cadence, and report that instead of your pulse.
- Skin and tissue differences: skin tone, tattoos, hair and blood flow to the skin can all change how much light comes back.
Software filters out much of this, but it has to guess, and guessing is hardest when heart rate changes quickly.
How validation studies judge a device
A quick guide to the numbers studies report:
- Concordance (rc): how closely two sets of readings agree, from 0 to 1. An rc of 1 is perfect agreement. The 2017 study below treated anything above 0.80 as acceptable.
- Mean bias: the average difference between the device and the reference. A bias near zero means errors cancel out on average, but says nothing about how big individual errors are.
- Limits of agreement: the range within which most individual differences fall. This is often the more useful number for an athlete, because it tells you how far off one reading could be.
A device can have a small average bias and still give single readings 20 beats a minute out.
The key studies
Fifty adults on a treadmill, a bike and a cross-trainer (2017)
The most widely cited study comes from Gillinov and colleagues at the Cleveland Clinic, published in Medicine & Science in Sports & Exercise in 2017. Fifty healthy adults (average age 38, 54% women) exercised on a treadmill, a stationary bike and an elliptical cross-trainer at rest, light, moderate and vigorous intensity. Each wore an ECG, an electrode chest strap, a forearm optical band and two randomly assigned wrist-worn devices from four common brands of the time.
- The chest strap agreed with the ECG almost perfectly across all conditions (rc = 0.996).
- The best wrist device reached rc = 0.92; two others came in at 0.83 and 0.81, and the forearm band and one wrist device were less accurate (0.75 and 0.67).
- On the treadmill, most optical devices performed well (rc 0.88 to 0.93), with one exception.
- On the bike, only three of the five optical devices reached acceptable agreement (rc above 0.80).
- On the cross-trainer with moving arm levers, none of them was accurate.
The authors concluded that accuracy depends on the type of exercise and that electrode chest straps should be used "when accurate HR measurement is imperative".
Fifty athletic adults running faster (2019)
A follow-up by Pasadyn and colleagues in Cardiovascular Diagnosis and Therapy in 2019 asked whether wrist sensors cope with harder running. Fifty healthy, athletic adults (68% men, average age 29) ran for two minutes at each of six treadmill speeds, from 4 to 9 miles an hour, wearing an ECG, a chest strap and two of four wrist devices.
- The chest strap again agreed best with the ECG (rc = 0.98).
- The best wrist device reached rc = 0.96, and the other three came in at 0.89.
So in running, the better wrist devices of 2017 to 2018 were close to the strap, and the weaker ones noticeably less precise.
A meta-analysis of 44 studies (2020)
A single study can be skewed by the devices it happened to test. A 2020 systematic review and meta-analysis by Zhang and colleagues in the Journal of Sports Sciences pooled 44 studies, covering 738 comparisons across 15 brands, against an ECG or chest strap.
- At rest, the average difference was essentially zero (-0.01 beats a minute).
- During treadmill walking and running, wrist devices read on average 0.51 beats a minute low, a trivial difference.
- During cycling, they read on average 4.55 beats a minute low.
- During resistance training, they read on average 7.26 beats a minute low, and the error grew by about 3 beats a minute for every 10 beat rise in heart rate.
The authors judged wrist devices acceptable for most common activities. For a triathlete, the split matters: running is the best case and the bike noticeably worse.
Sprints and short efforts (2021)
Most early studies used steady stages. Muggeridge and colleagues, in JMIR mHealth and uHealth in 2021, added sprints. Twenty healthy adults (9 women, 11 men, average age 40) did rest, steady cycling and a treadmill test to exhaustion, then four 15-second maximal bike sprints and four 30 to 50 m sprints on a non-motorised treadmill. An upper-arm optical sensor and a wrist-worn fitness band were compared with an electrode chest strap.
- The upper-arm optical sensor agreed well overall: average bias of -1 beat a minute, with limits of agreement from -20 to 19 beats a minute.
- The wrist band underestimated heart rate by 7 beats a minute on average, with limits of agreement from -46 to 33 beats a minute.
- Both were worse in the sprint session than in the steady trials, and the wrist band was particularly poor on the bike.
Inference: an optical sensor worn higher on the arm, where there is less wrist movement, may cope better than one on the wrist. This study tested one model of each, so it doesn't prove that every arm band beats every watch.
Skin tone and activity (2020)
The Bent study mentioned above tested consumer and research-grade wearables in 53 adults (32 women, 21 men, aged 18 to 54) spread evenly across the full range of skin tones, at rest, during breathing exercises, walking and typing.
- There was no statistically significant difference in accuracy across skin tones in this sample.
- Absolute error during activity was on average about 30% higher than at rest.
- Devices differed in how well they responded to changes in activity, even when they were accurate during steady states.
Inference: that fits what athletes see, a watch that is accurate once heart rate settles but lags when it rises or falls quickly.
How good is the chest strap itself? (2019)
If a strap is the reference, it's fair to ask how reliable it is. Gilgen-Ammann and colleagues, in the European Journal of Applied Physiology in 2019, had 10 healthy adults do five activities from low to high intensity while wearing a medical Holter ECG and a consumer chest strap. Both were compared with a visual check of the raw ECG.
- The chest strap detected 99.6% of heartbeats correctly overall, and 99.4% during high-intensity activities.
- The Holter monitor dropped to 89.8% during the high-intensity activities.
The sample was tiny, but it suggests a well-fitted strap is a sound reference during exercise. Straps aren't foolproof in everyday use. Inference: a strap that is loose or has dry electrodes at the start of a session tends to produce obvious dropouts or spikes, which are easier to spot than the subtle drift of an optical sensor.
Why the bike and intervals are harder
The studies agree on where optical sensors struggle. The reasons are partly measured and partly inference.
- Gripping and wrist angle. On the bike your hands grip the bars and your wrists are bent and braced. Inference: this changes blood flow and tissue tension under the sensor, which may explain why cycling errors were bigger than running errors in both the Gillinov study and the Zhang meta-analysis.
- Rapid changes. Optical software smooths the signal to filter out noise, which means it can lag behind real changes. In the Bent study, error rose during activity and devices differed in how they handled transitions. In a 30-second interval, a lag of even a few seconds is a large part of the effort.
- Cadence lock. Running cadence often sits in the same range as heart rate. If the sensor latches on to arm swing, you see a reading that matches your step rate. Inference: this is a likely cause of the classic early-run spike, when heart rate is still low and cadence is already high.
- Cold. Blood flow to the skin falls in the cold. No study above tested cold conditions, so poor readings on winter rides are plausible but unmeasured.
Swimming is a separate problem; see TriForward's "Swim Watches and Trackers: How Accurate Are Lengths, Distance and Wrist Heart Rate in Water?".
Irregular rhythms are a special case
All the studies above recruited people with a normal heart rhythm. A 2020 study by Al-Kaisey and colleagues in the International Journal of Cardiology looked at what happens when the rhythm is irregular. Thirty-two adults referred for 24-hour heart monitoring wore a smartwatch alongside the clinical monitor; most had persistent atrial fibrillation.
- In normal rhythm, the watches agreed well (bias under one beat a minute).
- In atrial fibrillation, they underestimated heart rate, and the error grew with heart rate: by 28 beats a minute at 100 to 120, 48 at 120 to 140 and 69 above 140.
This is a patient study, not an athlete study, but the message matters for masters triathletes, because atrial fibrillation becomes more common with age. A wrist reading that looks calm doesn't rule out an irregular rhythm, and a rhythm alert or symptoms belong with a doctor (see the safety notes).
Who was studied
- Sex split: most samples were reasonably balanced (54% women in Gillinov, 32 of 53 in Bent, 9 of 20 in Muggeridge), though Pasadyn's athletic sample was 68% men. No study analysed accuracy by sex.
- Fitness: most participants were healthy adults, not trained endurance athletes. Only Pasadyn recruited athletic adults, and only for treadmill running.
- Sessions: stages lasted a few minutes. None tested a long ride, a brick or a race.
- Devices: every device tested has since been replaced, and newer software may reduce error, but the evidence lags behind the products.
Does a few beats a minute matter?
It depends on what you use heart rate for.
For an easy run or ride, an error of a few beats makes little difference. Inference: if your easy ceiling is 145 beats a minute and the watch reads 3 beats low, you'd ride at 148 instead, which is still easy for most people.
For zone-based threshold work the margins are tighter. Inference: if your threshold heart rate is about 160 beats a minute, the meta-analysis average cycling error of 4.55 beats is close to 3% of that number, which can be the difference between a session at threshold and a session just under it. Individual readings can be much further out than the average, as the limits of agreement in the Muggeridge study showed.
For short intervals heart rate is a poor guide anyway, because it takes time to rise, so even a perfect sensor lags your effort. Many athletes pace these by power, pace or feel. TriForward's "Racing by Feel vs. Racing by Numbers" and "Do You Need a Bike Power Meter? Accuracy, Single- vs Dual-Sided and What It Changes", cover those alternatives.
How to check your own device
The INTERLIVE Network, a European university group, published an expert statement in the British Journal of Sports Medicine in 2021 (Mühlen and colleagues), based on 44 eligible studies, on how optical devices should be validated: against a proper criterion device, across realistic activities and intensities. You can borrow the same principles.
- Use a chest strap as your reference. Wear the strap and watch together, recording to separate devices or apps if you can.
- Test the sessions you actually do: a steady 40-minute run, a ride with some climbing or harder efforts, and one interval session.
- Overlay the two traces in your training software. Look for three things: an early spike in the first 10 minutes, a lag when intervals start and finish, and a steady offset on the bike.
- Refit before you judge. Inference: because movement of the sensor on the skin is a main source of error, a loose watch is the most fixable problem. Wear it snugly, slightly above the wrist bone, during sessions and loosen it afterwards.
- Recheck after major software updates.
If the watch tracks the strap on steady running but drifts on the bike or in intervals, use it for easy work and wear a strap when the numbers matter.
When the chest strap earns its place
The strap matters most for:
- Cycling, where wrist error was largest outside the gym.
- Interval and threshold sessions paced or judged by heart rate.
- Lab or field tests you use to set zones, including threshold tests.
- Heart rate variability or beat-to-beat data, where the strap detected 99.6% of beats correctly.
- Anyone whose watch fails the check above.
For steady, easy running, the wrist is usually close enough. An upper-arm optical band may be a middle ground, though the evidence is one small study.
A triathlon scenario
This example is illustrative, not from a study. A 42-year-old training for a middle-distance race uses wrist heart rate for everything. Her turbo threshold rides show heart rate 6 to 8 beats lower than she expects, and early-run readings jump to 170.
She wears a borrowed chest strap with her watch for a week:
- On easy runs, the traces overlap after the first 8 minutes. Before that, the watch shows a spike that matches her running cadence.
- On the turbo, the watch reads consistently low during the harder blocks, which fits the meta-analysis average for cycling.
- In a set of 2-minute run intervals, the watch lags at the start of each rep and overshoots after.
She keeps the watch for easy running, ignores its first 10 minutes, and wears a strap for bike sessions and intervals.
Common mistakes
- Trusting the first 10 minutes of a run, when spikes may be cadence lock.
- Assuming the bike is as accurate as the run.
- Pacing short intervals by heart rate from any sensor.
- Reading old studies as today's ranking.
- Ignoring symptoms because the number looks normal.
What we don't know
- How current devices perform. Published studies trail the products by several years.
- Accuracy in trained triathletes over long sessions, including long rides, bricks and races.
- The effect of cold, sweat and sunscreen on optical readings outdoors.
- Differences by sex, age and skin tone in athletes. One study found no skin-tone effect at rest and light activity, but it wasn't designed around hard exercise.
- Whether upper-arm optical bands are generally better than wrist sensors, beyond one small study.
- Whether better heart rate accuracy changes training outcomes. No study has tested that.
References
- Gillinov S, Etiwy M, Wang R, Blackburn G, Phelan D, Gillinov AM, Houghtaling P, Javadikasgari H, Desai MY. Variable accuracy of wearable heart rate monitors during aerobic exercise. Medicine & Science in Sports & Exercise, 2017;49(8):1697–1703. doi:10.1249/MSS.0000000000001284
- Pasadyn SR, Soudan M, Gillinov M, Houghtaling P, Phelan D, Gillinov N, Bittel B, Desai MY. Accuracy of commercially available heart rate monitors in athletes: a prospective study. Cardiovascular Diagnosis and Therapy, 2019;9(4):379–385. doi:10.21037/cdt.2019.06.05
- Zhang Y, Weaver RG, Armstrong B, Burkart S, Zhang S, Beets MW. Validity of wrist-worn photoplethysmography devices to measure heart rate: a systematic review and meta-analysis. Journal of Sports Sciences, 2020;38(17):2021–2034. doi:10.1080/02640414.2020.1767348
- Muggeridge DJ, Hickson K, Davies AV, Giggins OM, Megson IL, Gorely T, Crabtree DR. Measurement of heart rate using the Polar OH1 and Fitbit Charge 3 wearable devices in healthy adults during light, moderate, vigorous, and sprint-based exercise: validation study. JMIR mHealth and uHealth, 2021;9(3):e25313. doi:10.2196/25313
- Bent B, Goldstein BA, Kibbe WA, Dunn JP. Investigating sources of inaccuracy in wearable optical heart rate sensors. npj Digital Medicine, 2020;3:18. doi:10.1038/s41746-020-0226-6
- Gilgen-Ammann R, Schweizer T, Wyss T. RR interval signal quality of a heart rate monitor and an ECG Holter at rest and during exercise. European Journal of Applied Physiology, 2019;119(7):1525–1532. doi:10.1007/s00421-019-04142-5
- Al-Kaisey AM, Koshy AN, Ha FJ, Spencer R, Toner L, Sajeev JK, Teh AW, Farouque O, Lim HS. Accuracy of wrist-worn heart rate monitors for rate control assessment in atrial fibrillation. International Journal of Cardiology, 2020;300:161–164. doi:10.1016/j.ijcard.2019.11.120
- Mühlen JM, Stang J, Lykke Skovgaard E, Judice PB, Molina-Garcia P, Johnston W, Sardinha LB, Ortega FB, Caulfield B, Bloch W, Cheng S, Ekelund U, Brønd JC, Grøntved A, Schumann M. Recommendations for determining the validity of consumer wearable heart rate devices: expert statement and checklist of the INTERLIVE Network. British Journal of Sports Medicine, 2021;55(14):767–779. doi:10.1136/bjsports-2020-103148
Frequently asked questions
Is wrist heart rate accurate enough for running?
For steady running, mostly yes. A 2020 meta-analysis of 44 studies found wrist devices were within about half a beat a minute of the reference on the treadmill, on average. Individual readings can still be off, especially at the start of a run and during short, sharp efforts.
Why does my watch read oddly on the bike?
In a 2017 study of 50 adults, only three of five optical devices agreed acceptably with an ECG while cycling, and a 2020 meta-analysis found wrist devices read about 4.6 beats a minute low on the bike on average. Gripping the bars and a bent wrist are the likely culprits, although that's inference.
Do I need a chest strap for interval sessions?
If you use heart rate to judge short intervals, a strap helps. Electrode straps agreed with an ECG almost perfectly in several studies, while optical sensors were less accurate in sprints. For long, steady sessions the difference matters much less, and many athletes pace short intervals by power, pace or effort anyway.
How can I check my own watch?
Wear a chest strap and the watch together for a few typical sessions: a steady run, a ride and an interval set. Compare the two traces. If they overlap on steady efforts but the watch lags or spikes in intervals, use the watch for easy work and the strap when the numbers matter.
The evidence behind this guide
1 graded principle, 8 sources. Open one to see the sources.
Wrist optical heart rate is close to a chest strap during steady running but less reliable while cycling and in sprints and short efforts, where an electrode strap remains the better referenceModerate evidence8 sources
In practice: Use wrist heart rate for steady, easy running; use an electrode chest strap for cycling, intervals, zone tests and beat-to-beat data, or check your own watch against a strap on typical sessions. Device studies date quickly, and no study tested trained triathletes over long sessions. A wearable is not a medical device: wrist sensors underestimated heart rate in atrial fibrillation (Al-Kaisey 2020, 32 patients), so symptoms or rhythm alerts need a doctor.
Against ECG across treadmill, bike and elliptical at rest to vigorous intensity: electrode chest strap rc = 0.996; best wrist device rc = 0.92, others 0.83 and 0.81, a forearm optical band 0.75 and one wrist device 0.67. Most devices rc 0.88-0.93 on the treadmill; only three of five optical devices acceptable (rc > 0.80) cycling; none accurate on the elliptical with arm levers. Devices of the time; healthy adults, not athletes; short stages.
Treadmill running at 4-9 mph (2 min per stage) against ECG: chest strap rc = 0.98; best wrist device rc = 0.96; three other wrist devices rc = 0.89. Running only; 2017-2018 devices.
Mean difference versus criterion: -0.01 bpm at rest, -0.40 bpm in sleep, -0.51 bpm during treadmill walking and running, -4.55 bpm cycling and -7.26 bpm in resistance training, with error rising about 3 bpm per 10 bpm increase in heart rate during resistance training. Acceptable validity for most common activities; mostly non-athlete samples and older devices.
Against an electrode chest strap: an upper-arm optical sensor had bias -1 bpm (limits of agreement -20 to 19, r = 0.95); a wrist band underestimated by 7 bpm (limits -46 to 33, r = 0.8). Both were less valid during maximal bike and treadmill sprints than steady exercise; the wrist band was particularly poor cycling. One model of each device type.
No statistically significant difference in optical heart rate accuracy across skin tones; significant differences between devices and activity types, with absolute error during activity on average about 30% higher than at rest; devices differed in responding to changes in activity. Rest, breathing, walking and typing, not hard exercise.
Consumer chest strap correctly detected 99.6% of RR intervals overall and 99.4% during high-intensity activities, versus 94.6% and 89.8% for a Holter ECG; correlation r = 0.997. Very small sample.
Smartwatches agreed with Holter in sinus rhythm (bias < 1 beat) but underestimated heart rate in atrial fibrillation (bias -9 beats overall), worsening at higher rates: -28 at 100-120, -48 at 120-140 and -69 above 140 bpm. Patients, not athletes.
Best-practice validation framework for PPG heart rate wearables across six domains (target population, criterion measure, index measure, testing conditions, data processing, statistics), with checklists; calls for testing against a proper criterion across realistic activities and intensities.
Get The Forward
Research-led triathlon guidance, straight to your inbox. No spam, no filler — unsubscribe any time.
