How smartwatch sensors work, and what changes under the wrist

The question arrives within a minute of anyone seeing a smartwatch worn face down under the wrist: does the heart rate still work down there? It is a fair question, and it deserves better than the two usual answers, which are a cheerful yes from people selling bracelets and a knowing no from people who have never looked at how the sensor works. The honest answer has three parts. The physics of the sensor does not depend on which way the watch faces. Several things that matter far more than orientation can still spoil a reading. And the company that makes the most popular smartwatch says, in writing, that its sensors are meant for the top of the wrist. All three are true at once.

What the light sensor actually does

The green lights on the back of a smartwatch are photoplethysmography, PPG for short. LEDs send light into the skin, a photodiode a few millimetres away measures how much comes back, and because blood absorbs green light, the amount returning rises and falls with each pulse. The watch counts the rhythm of that flicker.

The word that matters is reflectance. The light goes in and comes back out through the same surface, so the emitter and the detector sit side by side on the case back. Nothing in that loop knows where the sky is. A watch on the top of the wrist and a watch on the underside both shine light into living tissue and read what bounces back.

That is the strong version of the argument, and it is correct as far as it goes. It just does not go all the way, because the tissue under each side of the wrist is not the same, and because orientation was never the main source of error in the first place.

What really spoils a reading

Motion comes first. A team in Germany recorded 24 hours of everyday life with a smartwatch and measured how much of the PPG signal an algorithm could actually analyse. On average it was 71.5% of the day, with a range from 34% to 100% between people, and above 94% between midnight and 6 am, when the wrist barely moves (Merschel et al., JMIR Formative Research, 2022). The more the wrist accelerated, the less usable signal there was.

Posture comes second, and it is less obvious. A 2025 study led by Peter Charlton in PLOS Digital Health found the best signal quality lying down and the worst standing with the arm hanging by the side, a drop in signal-to-noise ratio from 18.6 dB to 9.0 dB. Holding the sensor at heart height helped (Charlton et al., PLOS Digital Health, 2025). Detail worth keeping: that study recorded on the anterior side of the forearm, the inner side.

Then the skin itself. Apple's own guidance names tattoos whose ink can block the light, and cold weather that lowers blood flow in the wrist enough to stop a reading. It also notes that rhythmic movement such as running or cycling reads better than irregular movement such as tennis or boxing.

• Motion of the wrist, by far the biggest factor in everyday wear.
• Arm position relative to the heart.
• Fit: a sensor that lifts off the skin reads air.
• Skin: tattoos, cold, low perfusion.

Notice what is missing from that list. Which way the watch faces.

What Apple says about the wrist

Here is the part a bracelet maker would rather skip. On its page about getting the most accurate measurements, Apple writes: "In addition, the sensors will work only if you wear your Apple Watch on the top of your wrist." The same page asks for a fit that is "not too tight, not too loose", suggests tightening the band for workouts, and says the heart rate sensor should stay close to your skin (Apple Support).

Apple does not explain the sentence, and we will not pretend to explain it for them. The underside of the wrist has thinner skin, tendons close to the surface and the radial artery nearby, so the optical environment is different from the top. Different is not the same as worse, but it is not the same as tested either. We have not found a peer-reviewed study that compares a consumer smartwatch on the top and the underside of the same wrist, and the published validation work on these watches is done the way the manufacturer specifies.

Motion sensors and the ECG

The accelerometer and the gyroscope are the easy part. They are three-axis sensors, built to measure acceleration and rotation in every direction, which is how a watch recognises the pattern of a step or of a fall whatever its angle. Turning the watch face down changes the sign of a few numbers, not what the sensor can see.

The ECG is a different instrument altogether. On an Apple Watch, electrodes in the back crystal and the Digital Crown close an electrical circuit when you touch the crown with a finger of the other hand (Apple Newsroom). You do not press it. The recording takes 30 seconds, with your arms resting on a table or in your lap, and Apple asks that the watch be snug and on the wrist selected in the Watch app (Apple Support). We have a separate guide on reading those results.

With the watch under the wrist, reaching the crown with the other hand is a slightly different gesture. The instruction that matters does not change: if you are taking an ECG because a doctor asked for one, follow Apple's protocol to the letter.

How accurate these watches are, as worn in studies

The best evidence comes from pooled studies. A living systematic review of the Apple Watch in npj Digital Medicine found heart rate within a mean bias of -0.27 beats per minute against reference devices, with limits of agreement from -7.19 to 6.64. For atrial fibrillation it reported a specificity of 0.91 and a sensitivity of 0.79. Blood oxygen showed wide limits of agreement, and energy expenditure errors were frequently large (Lambe et al., npj Digital Medicine, 2026).

A 2025 meta-analysis in JACC Advances, across 26 studies and 17,349 patients, put pooled smartwatch detection of atrial fibrillation at 95% sensitivity and 97% specificity, with photoplethysmography and ECG-based methods performing comparably (Barrera et al., JACC Advances, 2025).

Two things follow. Heart rate at rest is very good, energy burned is a rough guess, and the gap between them has nothing to do with wrist position. And none of these numbers come from watches worn underneath, so they describe the ceiling, not a guarantee for every way of wearing one.

Where the Smartlet fits in this

The most common reason a smartwatch reads nothing is that it is in a drawer. A collector puts on a mechanical watch in the morning and the health data stops for the day. The Smartlet is a metal link bracelet that holds both: the mechanical watch on top, its watch adapters seated in the lug holes in place of the spring bars, and the smartwatch underneath, in the central part of the bracelet, against the skin. The bracelet is sized to the wrist by its links, so it sits snug, which is the one fit condition every source above agrees on (setup guide).

What the Smartlet does not do is change the sensor, its algorithm or Apple's specification. It changes how many hours the watch is worn at all. For background heart rate, sleep and the slow trends that only exist when the data has no gaps, those hours are the whole game. For a reading with medical consequences, put the smartwatch back on its own band, on top of the wrist, and take it the way the manufacturer says.

Nights are simple: the Clasp accessory closes the bracelet in place of the mechanical watch, and the smartwatch keeps tracking sleep. For training, the smartwatch goes back on its sport band, because no metal bracelet is made for sport.

Compatible smartwatches are listed in the smartwatch compatibility index, with dedicated pages for the Apple Watch, the Galaxy Watch and the Withings ScanWatch. For the mechanical side, start with watch brand compatibility. For the other ways to add sensors to a mechanical watch, see can you make an analog watch smart?.

This article explains how consumer smartwatch sensors work. It is not medical advice. Smartwatch readings are not a diagnosis, and any symptom or alert should be discussed with a doctor.

Smartlet, Bronze Medal at Concours Lepine 2025 and shown at CES 2026, keeps both watches on one wrist.