Your watch detects deep sleep by combining two main signals: how much you’re moving and what’s happening with your heart. A green light sensor on the back of the watch reads your pulse dozens of times per second, while a tiny motion sensor tracks even the smallest shifts of your wrist. Together, these signals create a fingerprint for each sleep stage, and deep sleep has the most distinctive pattern of all.
The Green Light on Your Wrist
That green glow on the underside of your watch is a sensor called a photoplethysmograph. It shines light into your skin and measures how much bounces back. With every heartbeat, blood pulses through the tiny vessels in your wrist, absorbing slightly more light. By tracking these fluctuations, the watch calculates your heart rate and, more importantly, the variation between beats.
This beat-to-beat variation is the key to identifying deep sleep. When you enter the deepest stage of sleep (called N3 or slow-wave sleep), your body’s “rest and digest” nervous system takes over. Sympathetic activity, the fight-or-flight side, quiets down. Your heart rate drops and stabilizes, blood pressure falls by roughly 10% compared to waking levels, and your arteries relax. The pulse signal your watch picks up changes shape: the peaks broaden, the timing between features of each pulse wave stretches out, and measures of arterial stiffness hit their lowest point of the entire night. All of these shifts are most pronounced during deep sleep compared to any other stage.
Heart rate variability, the subtle fluctuation in the gap between heartbeats, reaches its highest values during deep sleep. This reflects the surge in parasympathetic nervous system activity. Your watch uses this as one of its strongest clues: a sustained period of high heart rate variability, combined with a slow, steady heart rate, points toward deep sleep rather than light sleep or REM.
Why Stillness Alone Isn’t Enough
Your watch also contains a three-axis accelerometer, a chip that detects motion in every direction. During deep sleep, muscle tone drops and you barely move. The accelerometer picks up this near-total stillness. But here’s the problem: you can also be very still during other sleep stages, or even while lying awake. Motion data alone gets sleep staging right only about 57% of the time. When researchers tested models using just accelerometer data, accuracy was mediocre. The real improvement comes from combining motion with heart rate data, which roughly doubles the accuracy score.
So your watch doesn’t simply label every motionless stretch as deep sleep. It layers stillness on top of the cardiac signals. A long, quiet period with high heart rate variability and a slow, stable pulse gets flagged as deep sleep. A quiet period with more erratic heart rhythms and occasional eye-movement-related micro-shifts might get classified as REM instead.
Breathing Adds Another Layer
Some watches also estimate your breathing rate, either from the pulse signal itself or from subtle chest movements detected by the accelerometer. During deep sleep, breathing becomes remarkably regular. Researchers measuring breath-to-breath variability found it was lowest during N3, significantly lower than during REM or lighter sleep stages. The actual breathing rate doesn’t change much between stages (hovering around 15 to 16 breaths per minute), but the consistency of each breath is a useful marker. A stretch of clockwork-steady breathing reinforces the watch’s confidence that you’re in deep sleep.
How the Algorithm Puts It Together
Your watch doesn’t use a simple checklist. Modern sleep-staging algorithms rely on machine learning models, often a type of neural network, trained on thousands of nights of data. In one approach, researchers extracted over 130 features from heart rate variability and motion data, then fed them into a model that learns the patterns associated with each sleep stage. Another method uses a specialized network designed to process the raw accelerometer signal alongside handcrafted heart rate features, letting the model discover patterns a human engineer might miss.
These models also factor in timing. Deep sleep is most common in the first half of the night, and the algorithms know this. Your body’s circadian rhythm influences when deep sleep typically occurs, so the watch uses time-of-night as an additional input. Researchers have found that adding circadian timing features improves accuracy more than adding heart rate data on top of motion alone.
The watch analyzes your night in 30-second chunks called epochs. For each epoch, the algorithm assigns the most likely stage: wake, light sleep, deep sleep, or REM. The result is the colorful sleep chart you see in the morning.
How Accurate Is the Deep Sleep Reading
Compared to polysomnography, the clinical gold standard that uses brain wave electrodes, wearables do a reasonable job but aren’t perfect. A 2024 study comparing three popular devices against polysomnography found overall four-stage agreement rates of 76.3% for Oura Ring, 75.0% for Apple Watch, and 70.9% for Fitbit. For deep sleep specifically, the devices varied more. Oura correctly identified about 80% of the deep sleep epochs that polysomnography confirmed. Apple Watch caught only about 51%, and Fitbit landed at 62%.
The practical impact shows up in your nightly totals. Oura’s deep sleep estimates were statistically indistinguishable from the lab results. Fitbit underestimated deep sleep by about 15 minutes per night, while Apple Watch underestimated it by a striking 43 minutes, misclassifying much of it as light sleep. These differences matter if you’re comparing yourself to population averages, but they’re less important if you’re tracking your own trends over time. A device that consistently underestimates by the same amount still shows you which nights were better or worse.
What Can Throw It Off
Several factors can cause your watch to misread deep sleep. A loose-fitting watch lets the optical sensor bounce, creating noisy pulse data. The algorithms used by some devices filter out temperature readings outside a normal skin range (roughly 31 to 40 degrees Celsius), so a very cold wrist or heavy blanket trapping heat can introduce errors. Skin tone also matters: research has shown that the green light sensor produces varied signal quality in people with darker skin, which can affect the reliability of heart rate variability readings.
Alcohol, medications, and sleep disorders can also change the relationship between your heart signals and your actual brain state. A watch trained on healthy sleepers may misinterpret the heart rate patterns of someone with sleep apnea or someone who had several drinks before bed. The cardiac signature of deep sleep in those situations may not match the patterns the algorithm learned.
Different Watches, Different Approaches
Not all watches weigh their sensors the same way. Apple’s sleep staging algorithm relies primarily on accelerometer patterns, supplemented by heart rate data. Fitbit and Google Pixel watches lean more heavily on heart rate variability and have shown somewhat better agreement with polysomnography for slow-wave and REM detection. Garmin devices follow a similar sensor fusion approach, though independent validation shows mixed results for staging accuracy. The Oura Ring, worn on the finger where arterial signal is stronger, has shown the closest agreement with lab measurements for deep sleep duration.
Regardless of brand, the underlying principle is the same: your watch reads the physical echoes of what your brain is doing. Deep sleep produces a unique combination of near-total stillness, a slow and steady heart, high beat-to-beat variability, relaxed blood vessels, and metronomic breathing. No single signal is definitive, but layered together, they give your watch a surprisingly good read on what’s happening inside your sleeping brain.