Your smartwatch’s optical heart rate sensor is essentially a high-speed flashlight and a camera aimed at your skin. The green LEDs blast photons into your capillaries, and a photodiode counts how many bounce back—a technique called photoplethysmography (PPG). It’s clever, but it’s not a medical-grade electrocardiogram. The gap between what a wrist-worn device measures and what a hospital machine reports can be significant, especially during high-intensity intervals or for people with darker skin tones or tattoos. Yet millions of users now rely on these numbers to guide their training, sleep, and stress management. This guide will walk you through the most common fitness metrics—heart rate, SpO₂, VO₂ max, sleep staging, and stress—with hard data on accuracy, specific sensor hardware, and honest limitations. You’ll learn which numbers are clinically useful, which are marketing fluff, and how to interpret your watch’s readings without fooling yourself.
The PPG sensor inside most smartwatches—like the Texas Instruments AFE4900 analog front-end used in many Samsung Galaxy Watches and the custom Apple-designed sensor in the Series 8 and Ultra—works by measuring changes in blood volume. During rest, the error rate compared to a chest strap (which uses electrical signals) is typically 2–5%. A 2021 study by Shcherbina et al. found the Apple Watch Series 6 to have a mean absolute error of just 2.5% for heart rate during treadmill walking, but that error jumped to 9.6% during high-intensity running. The Garmin Forerunner 265, using a similar PPG but with a different algorithm (Firstbeat, now owned by Garmin), showed slightly better accuracy during intervals, with a 7.1% error in the same study.
Why does this matter? If you’re using heart rate zones for training (e.g., Zone 2 for endurance), a 10% error can push you from a recovery pace into threshold work. The optical sensor also struggles with cadence locking—when your arm swing frequency matches your step rate, the watch can mistake foot strikes for heartbeats. This is especially common on treadmills. To mitigate this, many Garmin watches now include a “Cadence Lock” alert in the settings. For the most accurate data, especially during structured workouts, a chest strap (like the Polar H10) remains the gold standard, with a mean error under 1% even during sprints. But for daily wear and casual exercise, the wrist-based PPG is good enough for trend tracking—just don’t trust a single reading at peak effort.
Blood oxygen saturation (SpO₂) is measured using red and infrared LEDs. The watch shines both wavelengths through your skin; oxygenated hemoglobin absorbs more infrared, while deoxygenated hemoglobin absorbs more red light. The ratio gives a percentage, typically 95–100% for healthy individuals. The Apple Watch Series 6 and later use a second-generation photoplethysmograph with four LEDs, while the Garmin Fenix 7 uses a single red-IR pair. A 2021 validation study published in the Journal of Medical Internet Research compared the Apple Watch SpO₂ to a Masimo Rad-7 pulse oximeter and found a mean difference of 0.2% with 95% limits of agreement from −2.8% to 3.2%. That’s within the FDA’s requirement for over-the-counter pulse oximeters (±3%), but only for values above 90%.
Below 90%—the clinical threshold for hypoxemia—accuracy drops sharply. The same study reported a bias of −1.5% at saturations below 90%, meaning the watch tends to read lower than the medical device. This is dangerous if you’re relying on it to detect breathing problems. The sensor also fails during movement, low perfusion (cold hands), or with nail polish. The Garmin Venu 2 Plus, using the TI AFE4900, showed even wider limits of agreement: −4.1% to 4.8% in a small 2022 study. Smartwatch SpO₂ is useful for overnight tracking to spot trends (e.g., altitude acclimatization), but it is not a substitute for a medical pulse oximeter. The FDA has cleared the Apple Watch’s irregular rhythm notification for atrial fibrillation, but not its SpO₂ for any diagnosis. Treat it as a rough guide, not a clinical tool.
VO₂ max—the maximum amount of oxygen your body can use per minute per kilogram of body weight—is a strong predictor of endurance and overall mortality. Smartwatches estimate it using heart rate, pace, age, weight, and sometimes heart rate variability (HRV). The algorithm, typically from Firstbeat (now Garmin) or a similar licensed engine, assumes a linear relationship between heart rate and oxygen consumption. Direct measurement in a lab requires a mask and gas analyzers costing tens of thousands of dollars. The watch’s estimate is a statistical guess, and for most people it’s off by 10–15%. A 2020 study in Medicine & Science in Sports & Exercise found the Garmin Forerunner 945 underestimated VO₂ max by an average of 3.5 ml/kg/min in trained runners, while the Apple Watch Series 5 overestimated by 1.8 ml/kg/min.
To get a reliable estimate, you need a steady-state outdoor run of at least 10 minutes at a constant pace—treadmill runs don’t count because GPS is required. The watch also needs your correct maximum heart rate; if you haven’t set it manually, it uses the age-predicted formula (220 minus age), which can be off by 10–20 beats per minute for fit individuals. Some watches, like the Garmin Forerunner 965, allow you to enter a lab-measured VO₂ max as a calibration point. For the most accurate numbers, use a chest strap during the run; wrist-based heart rate errors can skew the estimate further. A typical sedentary adult has a VO₂ max around 35 ml/kg/min, an active person around 45, and an elite endurance athlete can exceed 70. The watch’s estimate is useful for tracking long-term trends (e.g., +2 ml/kg/min over 3 months), but don’t treat the absolute number as gospel.
Polysomnography (PSG) uses EEG, EOG, and EMG to classify sleep into wake, light, deep, and REM stages. A smartwatch replaces all that with an accelerometer, heart rate, HRV, and sometimes SpO₂. The agreement with PSG for sleep/wake detection is around 70–80%—meaning it correctly identifies whether you’re asleep or awake about three-quarters of the time. But for specific stages, the numbers are worse. A 2022 study on the Fitbit Sense (which uses a proprietary algorithm) showed sensitivity for sleep detection at 0.96 (excellent), but specificity for wake detection at just 0.61—meaning it often calls wakefulness sleep. For REM and deep sleep, the Cohen’s kappa (a measure of agreement beyond chance) was 0.42 and 0.38 respectively, which is considered “fair” at best.
The Garmin Fenix 7 uses a combination of accelerometry and HRV from the PPG sensor (the Bosch BHI260AP handles motion processing). In a 2023 preprint, its sleep stage classification accuracy was 72% for light sleep, 68% for deep, and 65% for REM compared to PSG. The Apple Watch, which uses a “sleep tracking” mode that relies on a combination of accelerometer and heart rate data, showed similar numbers in a 2021 study: overall stage accuracy of 71%. The practical takeaway: your watch is good at telling you when you fell asleep and woke up (within 10–20 minutes), but the breakdown into stages is a rough approximation. If you consistently see less than 30 minutes of deep sleep, that’s a useful trend—but don’t panic if one night shows zero deep sleep. The watch may have misclassified it.
Stress scores are almost entirely derived from heart rate variability (HRV)—specifically the RMSSD (root mean square of successive differences) of your R-R intervals. A high HRV indicates a relaxed, parasympathetic state; low HRV suggests stress or fatigue. Smartwatches measure HRV by analyzing the tiny variations in the time between heartbeats using the same PPG sensor. The problem is that PPG-based HRV is less accurate than ECG-based HRV, especially during movement. A 2022 study comparing the Apple Watch Series 7’s HRV (via the Breathe app) to a Polar H10 chest strap found a mean absolute error of 12 ms for RMSSD. Typical healthy adult RMSSD ranges from 40–60 ms, so a 12 ms error is significant.
Garmin’s Body Battery uses a proprietary algorithm combining HRV, activity, stress, and sleep to give a 0–100 energy score. It’s a useful proxy for recovery, but it’s not validated against any clinical measure. The same goes for Fitbit’s Stress Management Score, which adds self-reported mood. These scores can be gamed—if you sit still for 10 minutes, your stress score drops. They are best used as a trend over days, not a real-time diagnostic. The sensor hardware matters: the TI AFE4900 in many watches has a sampling rate of 64 Hz for PPG, which is sufficient for HRV analysis during rest but not during movement. For the most accurate HRV readings, take a 5-minute seated measurement first thing in the morning, before coffee. Some watches (e.g., Garmin) do this automatically during sleep. The numbers are useful for guiding training intensity—if your morning HRV is 20% below your baseline, take it easy that day.
Battery life is the single biggest practical difference between smartwatches, and it’s directly tied to how aggressively you use the health sensors. An Apple Watch Series 9 lasts about 18 hours with typical use (including one 60-minute GPS workout) and 6 hours with continuous GPS. The Samsung Galaxy Watch 6 Classic manages about 40 hours in the same scenario, but only 8 hours with GPS always on. In contrast, the Garmin Fenix 7 Solar can stretch to 18 days in smartwatch mode (with solar) and 57 hours in GPS mode—because its MIP (memory-in-pixel) display is always-on with minimal power, and it uses a lower GPS polling rate (1 Hz vs 1 second). The Forerunner 265, with an AMOLED screen, gets 13 days smartwatch
Honest reviews and the best value picks, tested by us.
Honest reviews and the best value picks, tested by us.
