Chest strap vs wrist heart rate: how accurate is your watch?

You are halfway up a hill with your watch reading 148 and you are fairly sure it is wrong. It feels like 170. Two minutes later it jumps to 176, exactly as you slow to a walk. The question that occurs to you on that hill is a fair one: how much should I trust this number?

The answer is neither "completely" nor "not at all". It depends on what your arm is doing. This page is about exactly that: how the two technologies measure, how large the error is in studies against ECG, and in which situations the difference actually changes how you train.

Two technologies measuring two different things

The first thing to grasp is that a watch and a strap are not measuring the same thing with different levels of precision. They are measuring different phenomena.

A chest strap holds electrodes against your chest and picks up the electrical discharge that makes the heart muscle contract. It is the same signal an ECG reads, with fewer leads. The strap therefore knows when the heart beat, not when the blood arrived.

A watch does the opposite. It shines light from small LEDs into the skin on the back of your wrist and measures how much comes back. When a pressure wave passes, the vessels under the sensor briefly fill with more blood, more light is absorbed, and the reflected signal dips. The method is called photoplethysmography. So a watch counts pulse waves out at the wrist, not heartbeats in the chest.

In practice the two almost always agree. But they can diverge, and they do so for entirely different reasons. A strap loses the signal when its electrodes dry out or when it slips. A watch loses it when the light path between LED and skin is disturbed, meaning when the sensor lifts, when the skin is compressed, or when blood flow at the wrist changes for reasons other than the heartbeat.

What the comparisons actually show

The starting point is a study from the Cleveland Clinic. Gillinov and colleagues had 50 healthy adults, mean age 38, exercise on a treadmill, a stationary bike and an elliptical trainer while wearing an ECG as the reference, a chest strap, an optical monitor on the forearm, and two randomly assigned watches, one on each wrist. The results appeared in Medicine & Science in Sports & Exercise in 2017.

Agreement with ECG was scored on a concordance measure where 1 means the two measurements are identical. Pooled across all exercise types, it looked like this:

The strap is essentially at ECG level, which is not surprising, because it is the same measurement principle. What is striking is how widely the optical monitors scatter, from 0.92 down to 0.67, despite being tested in the same study during the same kinds of session.

But an average across all exercise types hides the most important part of the finding. The same study broke the results down by activity:

Read that last row again. It is not that a cheap watch did worse than an expensive one. None of them coped. The same Apple Watch that scored 0.94 without arm levers dropped below 0.80 with them. The authors' own conclusion was that electrode-containing chest monitors should be used when accurate heart rate measurement is imperative.

The finding has also been tested in a more vulnerable group. The same research team measured 80 patients in cardiac rehabilitation and found that none of the watches tested matched the chest monitor. That paper has since had an erratum published, so no individual figures from it are quoted here.

It is not movement that breaks the reading, it is the wrong kind of movement

Here is the most important point on this page, and it runs against intuition. The naive explanation is that shaking disturbs an optical reading, so the error ought to grow with how much you move. The data does not support that.

On the treadmill, where the arm swings most, the optical monitors were at their best. On the elliptical with arm levers, where the hands hold onto something that is itself moving, they were at their worst. The bike, where the arms barely move at all, sat in between.

A second dataset points the same way. Reddy and colleagues had 20 healthy adults complete a maximal test, a free-weight resistance circuit, an interval session and ordinary daily activities, with a chest strap as the reference. Published in JMIR mHealth and uHealth in 2018.

Same intensity, same person, same watch. What changes is what the hands are doing. On the treadmill the error for one of the watches was 0.5 percent. On the bike the same watch was off by 14.3 percent, nearly thirty times as much. The authors state plainly what they saw: error was greater during high-intensity activities when repetitive wrist motion was absent, and when the exercise mode indicator was not used.

The explanation lies in how a watch actually arrives at a number. The optical signal is full of noise, so the algorithm uses the motion sensor to model the shaking and subtract it. An arm swinging in a running stride produces a regular, predictable motion that can be described mathematically and removed. A steady grip on a handlebar gives the algorithm nothing to subtract, while the forearm muscles tense, press on the tissue beneath the sensor, and change the blood flow at the exact spot being measured.

Even the manufacturer's own validation points in the same direction. In a 2024 validation of a modern wristband, the largest deviations under motion were in male participants, which the authors themselves attributed to stronger forearm contractions. That study was run by the company that makes the wristband, with every author a shareholder, so it should be read as the manufacturer's own account. The mechanism it describes, however, is the same one two independent groups arrived at by measurement.

The average is right, the individual number is not

The next point is the one most often missed, because it requires looking at a statistic other than the mean.

In a 2016 study, Jo and colleagues put 24 people through a 77-minute protocol that included cycling, walking and running as well as resisted arm raises, resisted lunges and an isometric plank. The reference was a 12-lead ECG. The better of the two watches read 2.5 beats low on average. That sounds excellent.

But the limits within which a single reading could fall ran from 19.3 beats too high to 24.4 beats too low. That is a spread nearly 44 beats wide. The weaker watch in the same study averaged 8.8 beats too low with a spread of 66 beats. Above 116 beats per minute, the weaker watch's spread widened to just over 82 beats.

The hardware dates from 2016 and today's sensors are better. The point is not the exact figures but the shape of the error, because the shape recurs in newer work. In the 2024 manufacturer study, the limits under motion ran from 8.05 beats too low to 9.06 beats too high. The spread has shrunk from roughly 44 beats to roughly 17. It has not disappeared.

In practice this means one thing, and it is worth carrying with you: an average across a whole session is usually right, while a single number you glance at mid-interval can be badly wrong. It is the same logic that governs resting heart rate, where a daily value is stable and a snapshot is not.

The error leans one way

In both Reddy's and Jo's measurements the deviation was predominantly negative. On average the watches read low, rather than scattering evenly high and low.

That matters for training, for two reasons that pull the same way: both nudge you into training harder than you intended.

During a hard session, underreading makes the effort look easier than it was. If you steer the session by heart rate, you risk pushing harder than planned, because the number never climbs to where you aimed. In interval training specifically that is a problem, because the margins are thinnest there. If you do your intervals on a bike or a rowing machine it gets worse, because the grip adds its own error on top.

During an easy session the same error works on you differently. If the watch reads low, you assume you have room left and pick up the pace. That is precisely the mistake that stops so many people ever finding their true easy level in zone 2. Going by feel and breathing rather than by the number is often more reliable than it sounds.

When the wrist is enough and when it is not

Read the table from the top and you will see that most everyday uses sit in the safe half. That is the honest picture. A middle-aged person who wants to follow their fitness across months, sleep better and run easy needs no chest strap. It is the person steering hard sessions by numbers, or training with their hands wrapped around something, who gets value from one.

Where the sensor sits matters

It is tempting to blame the technology. A smaller 2019 study points to the position instead. Bunn and colleagues had 22 recreationally active adults wear an optical monitor in an earbud and an optical watch at the same time, with a chest strap as the reference, across treadmill work, high-intensity intervals and outdoor activity. The earbud had a mean absolute percentage error of 3.14 percent and an agreement of 0.939. The watch came in at 5.73 percent and 0.771. Both are optical. What separates them is where they sat. The authors wrote that the watch's accuracy was likely affected by wrist movement during the interval session.

One caveat belongs here: position alone does not decide it. In Gillinov's study one of the optical monitors sat on the forearm, and pooled across all activities it came second from bottom of the optical monitors at 0.75. On the bike, though, it was one of the three that held up, alongside two of the wrist watches. So the ranking reshuffles by activity, which is the same point the rest of this page makes.

Getting more out of the watch you already own

Before you buy anything, there are moves that cost nothing. Two of them are directly supported by the measurements above.

• Select the activity on your watch. Reddy's group explicitly found larger errors when the exercise mode indicator was not used. The algorithm changes how it interprets motion based on what you have told it you are doing.
• Let the arm swing. Running with your hands on your hips, or holding the treadmill rail, removes exactly the rhythmic motion the algorithm relies on.
• Wear it snug, a little above the wrist bone. The sensor needs steady skin contact. A watch that shifts with every stride is partly measuring air.
• Judge the number against the effort. If the watch says 148 while you can barely speak, it is the watch that is wrong, not your body.
• Compare only against your own history from the same device. Two different monitors produce two different series, even when both are reasonable. The same goes for your watch's VO2 max figure, which is partly based on heart rate. How far out it can be is covered in the guide to what VO2 max is.

What a chest strap does not fix

A strap is not a guarantee, it is a better starting point. It carries its own weaknesses, and they are of a different kind: they concern handling rather than measurement principle.

The electrodes need to be moistened to conduct. A dry strap at the start of a session often produces implausible numbers for the first few minutes, which then settle on their own. The strap needs to sit firmly just below the chest muscles, because any slack breaks contact. It needs a battery or a charge. And it needs to be on your body, which is the commonest reason straps end up in a drawer: most people forget them.

Weigh that up and the advice comes out less dramatic than the headline suggests. A strap sitting at home measures worse than a watch on your arm.

The margin of error travels up the chain, too. If your device builds a daily score out of the night's measurements, that score inherits whatever error was in the raw material, which is one of the two reasons two devices can disagree about the same night. The whole question is covered in the guide to readiness scores.

Skin tone, sex, and who the measurement works less well for

Optical measurement depends on how light travels through skin, and skin varies. The question is not whether that affects the reading but by how much.

A 2026 study collected optical wrist readings from participants of differing skin tones and sexes across a range of activities. Baseline error was higher for participants with darker skin tones and for women. The skin-tone parity gap came to 1.2 beats per minute in mean absolute error, and it could be closed entirely using methods that discard uncertain readings. No equivalent figure is reported for sex.

That figure is worth reading in both directions. The gap is real and measurable, which means it belongs in the open rather than swept aside. It is also small next to the error the same technology makes when you grip a handlebar: 11.4 and 14.3 percent for the two watches in Reddy's data. The activity you are doing affects your reading more than your skin tone does.

When the number is a health signal rather than a training signal

There is one situation where the difference between the technologies stops being about training. That is when the heart is not beating regularly.

An optical monitor counts pulse waves. If the rhythm is irregular, those waves become irregular in both strength and spacing, which is harder to count correctly than a steady rhythm. A 2023 study compared an activity wristband against continuous ECG monitoring in 70 stroke patients with a mean age of 79.4 years. In normal rhythm, agreement was good at 0.791. During active atrial fibrillation it fell to 0.211, with a mean absolute percentage error of 16.48 percent.

The same study assessed the wristband's irregular rhythm notification. It caught 34 percent of those who genuinely had atrial fibrillation, while never alerting a single person who was in normal rhythm.

Two conclusions follow, and both are practical. A warning is worth taking seriously, because in this dataset the feature never raised a false alarm. An absent warning proves nothing, because the feature missed two cases in three. That patient group was considerably older and sicker than a healthy 45-year-old, so the figures do not transfer directly to your own watch. The direction is clear all the same: a watch can raise a suspicion, it can never rule one out.

No consumer product makes a diagnosis. If you feel irregular beats, or you have symptoms, an ECG is what gives you an answer.

Frequently asked questions

Is a wrist heart rate monitor accurate enough?

For most purposes, yes. When you are still or running on level ground, a modern watch sits close to an ECG. In Gillinov's 2017 comparison, every optical monitor but one landed between 0.88 and 0.93 for agreement with ECG on a treadmill, and the exception fell to 0.76. The problem is not the average level but the spread in individual readings, and the fact that the spread grows sharply the moment your hands grip something.

Do I need a chest strap?

Only if you steer your training by heart rate in a situation where the wrist is weak. That means cycling, spin classes, strength work, the rowing machine, and intervals where you are trying to hit a range. If you mostly run easy and walk, and you use heart rate for resting values and week-to-week trends, a strap adds almost nothing. Remember too that a strap only measures on the days you actually put it on.

Why does my watch show the wrong heart rate when I lift weights or cycle?

Because the error does not come from movement itself but from the wrong kind of movement. Your watch uses its motion sensor to subtract shaking from the optical signal, which works well when your arm swings rhythmically as it does when you run. Gripping a handlebar or a dumbbell gives the algorithm no rhythmic motion to subtract, while your forearm muscles tense and press against the sensor. In Reddy's measurements the error was largest during hard intervals on a bike and smallest during hard intervals on a treadmill.

Can I measure HRV with a watch instead of a chest strap?

At rest it often works. Heart rate variability asks more of a sensor than average heart rate does, because the measure is built on the gap between individual beats rather than on an average across a few seconds. In Holmes's 2020 study, where the optical reading was taken by a phone app at the fingertip rather than at the wrist, the measure sat close to ECG at rest but drifted clearly apart after a strength session: the difference from ECG grew from an effect size of 0.26 to 0.42 at rest up to 1.14 afterwards. In practice that means measuring at rest, lying still, in the same way every time, and comparing only against your own history from the same device. What the measure itself means is covered in the guide to heart rate variability.

Does it mean anything if my watch warns about an irregular heart rhythm?

A warning is worth taking seriously, but the absence of one proves nothing. In a study of 70 stroke patients with a mean age of 79.4 years, the feature caught 34 percent of those who genuinely had atrial fibrillation, while it never warned anyone who was in normal rhythm. During active fibrillation the watch's heart rate reading also dropped sharply in reliability. That patient group was considerably older and sicker than a healthy 45-year-old, so the figures do not transfer directly. The conclusion still holds: a watch can raise a suspicion, it cannot rule one out.

Sources for the figures above: Gillinov et al., Medicine & Science in Sports & Exercise 2017 (PMID 28709155); Etiwy et al., Cardiovascular Diagnosis and Therapy 2019 (PMID 31275816, with published erratum PMID 32695646); Reddy et al., JMIR mHealth and uHealth 2018 (PMID 30530451); Jo et al., Journal of Sports Science and Medicine 2016 (PMID 27803634); Bunn et al., International Journal of Exercise Science 2019 (PMID 30899350); Holmes et al., Sensors 2020 (PMID 33050249); Meza et al., Sensors 2023 (PMID 37430546); Ray, Collins and Ponnapalli, Physiological Measurement 2026 (PMID 41875538); Chen et al., Digital Biomarkers 2024 (PMID 39670276, the manufacturer's own validation); 1177, Sweden's national health guide, on palpitations and altered heart rhythm.