Bottom line: Your tracker is excellent at some things, mediocre at others, and close to making things up in one specific area. Heart rate at rest and steady effort is genuinely good. Sleep versus wake is good. Sleep stages are a coin flip dressed up as a pie chart. And calorie burn, the number most people make decisions from, was off by 27 to 93 percent in the largest lab comparison ever run. Here is what the validation research actually found, study by study, so you know which numbers to trust.
Every figure below comes from published peer-reviewed validation research comparing consumer devices against laboratory gold standards: ECG for heart rate, indirect calorimetry for energy expenditure, and polysomnography for sleep. We name the study, the sample size, and the funding source where it matters. This is a research summary, not a product test, which is the honest thing this site is built to do.
The Short Version, Ranked by How Much You Should Trust It
| What it measures | How good is it | Best evidence |
|---|---|---|
| Heart rate, resting and steady effort | Genuinely accurate | 6 of 7 devices under 5% error (Stanford, 2017) |
| Sleep vs wake | Genuinely accurate | Sensitivity 95%+ across devices (Sensors, 2024) |
| Heart rate during intervals | Degrades with intensity | Error rises as effort rises, multiple studies |
| Sleep stages (light, deep, REM) | Mediocre | Stage sensitivity 50 to 86% (Sensors, 2024) |
| Detecting quiet wakefulness | Poor | Specificity 29 to 52% (SLEEP Advances, 2025) |
| Calories burned | Do not trust it | 27 to 93% error, none acceptable (Stanford, 2017) |
That ordering is the single most useful thing on this page. The features marketed hardest are, broadly, the ones that hold up worst.
Heart Rate: The Thing They Do Well
The reference study here is from Stanford University School of Medicine, published in the Journal of Personalized Medicine (2017). Sixty volunteers wore seven devices while their heart rate and energy expenditure were measured against clinical instruments. Six of the seven devices measured heart rate with a median error under 5 percent. Apple Watch came in lowest at around 2 percent, Samsung Gear S2 highest at 6.8 percent.
For context on how far this came: one of the study authors described early wearables as effectively random number generators. Modern optical sensors are a real engineering achievement, and when you are sitting still or holding a steady pace, the number on your wrist is close to what an ECG would tell you.
Then it gets complicated. Optical sensors work by shining light into your skin and reading changes in blood flow, which means anything disturbing that optical path degrades the signal. Reviews across the literature find the same pattern repeatedly: accuracy falls as exercise intensity rises. During intervals your heart rate is changing fast, your wrist is moving hard, and tendons are shifting under the sensor. Some devices lag. Some latch onto your cadence and report your step rate as your pulse.
A systematic review covering 32 studies, 1,085 participants and 16 devices concluded that only certain wrist monitors have been validated to an acceptable standard, with mean percentage error running as high as 20 percent depending on the model. Brand and model matter more here than most buying guides admit.
Two more findings deserve more attention than they get. The Stanford team reported that skin tone and body mass index affected accuracy, with darker skin and higher BMI associated with greater error. That is a meaningful equity problem in a product category sold as universal, and it is still under-researched.
Calories: The Number You Should Stop Looking At
This is the finding worth remembering, and it comes from the same Stanford study, which makes the comparison clean. The researchers set an acceptable error threshold for energy expenditure at under 5 percent. Not one device met it.
The most accurate device tested, the Fitbit Surge, was off by a median of 27.4 percent. The least accurate, the PulseOn, was off by 92.6 percent. Read that again: the best performer in a controlled laboratory setting missed by more than a quarter, and the worst nearly doubled or halved reality depending on direction.
The reason is structural rather than a bug someone will patch. Your device knows your movement, your heart rate, and the profile data you typed in. It does not know your metabolic efficiency, your muscle mass, your mitochondrial density, or your training history. Two people of identical age, height, weight and sex running side by side genuinely burn different amounts of energy, and no wrist sensor can see the difference. Each manufacturer then applies a proprietary algorithm to guess, which is why two devices on the same person disagree.
The practical consequence is specific and it costs people results. If your watch says 600 calories and you eat back 600 calories, you may have just eaten back 800 or 400. This is the single most common way tracking undermines a fat loss effort, and it is why our smart scale guide argues for judging progress on weight trend and how your clothes fit rather than on a calorie ledger built from an estimate with a 27 percent floor on its error.
Sleep: Good at the Simple Question, Weak at the Interesting One
Sleep is where the research has moved fastest, and where the nuance actually matters.
The most cited recent validation is Robbins and colleagues, published in Sensors (October 2024), run at Brigham and Women’s Hospital. Thirty-five adults spent a night wearing an Oura Ring Gen3, Fitbit Sense 2 and Apple Watch Series 8 while monitored with polysomnography, the clinical gold standard that reads brain activity directly.
The good news is real. For simply detecting sleep versus wake, sensitivity was 95 percent or higher for all three devices, which beats many older research-grade actigraphy devices. If your tracker says you slept seven hours, that total is probably close.
The staging is a different story. For discriminating between light, deep and REM sleep, sensitivity ranged from 50 to 86 percent across devices: Oura 76.0 to 79.5 percent, Fitbit 61.7 to 78.0 percent, Apple 50.5 to 86.1 percent. Four-stage agreement measured by Cohen’s kappa came out at 0.65 for Oura, 0.60 for Apple, 0.55 for Fitbit. In plain terms, the coloured hypnogram in your app is a reasonable guess, not a measurement.
One disclosure that belongs in any honest summary: that study was funded by Oura, though it was independently designed and conducted by the hospital researchers. Oura also performed best in it. That does not make the findings wrong, and the numbers are consistent with the broader literature, but you should know it before you weight the result.
Which is why the second study matters more. A six-device validation published in SLEEP Advances (2025) tested Fitbit Sense, Fitbit Charge 5, Apple Watch Series 8, WHOOP 4.0, Withings ScanWatch and Garmin Vivosmart 4 against polysomnography. All detected over 90 percent of sleep epochs. But specificity ranged from just 29 to 52 percent, and Cohen’s kappa ran from 0.21 to 0.53, described as fair to moderate agreement.
That specificity number explains something you have probably experienced. Low specificity means the device over-labels sleep: it catches nearly all your real sleep, but it also scores quiet wakefulness as sleep. Lying still in the dark, awake and frustrated, frequently gets recorded as sleeping. This is the mechanism behind the common complaint that a tracker credits you with hours you know you spent staring at the ceiling.
How We Evaluate
We do not run laboratory validation and we are not a testing lab. This article is a synthesis of published peer-reviewed research, and every figure above is attributed to a named study with its sample size stated so you can check it: the Stanford energy expenditure and heart rate study in the Journal of Personalized Medicine (2017, 60 participants, 7 devices), the Robbins et al. sleep validation in Sensors (2024, 35 participants, 3 devices), Oura-funded and disclosed as such, and the six-device sleep validation in SLEEP Advances (2025). Where studies disagree we say so rather than picking the flattering one. Where we draw a practical conclusion from the evidence, that is our editorial judgment and it is labelled.
What to Actually Do With This
Trust the trends, not the absolute numbers. A device with a consistent 20 percent error is still useful for telling you whether this week was harder than last week, because the error is roughly consistent for you. It is useless for telling you a true value. Almost every good use of wearable data is comparative.
Never eat back exercise calories from a wrist estimate. The error range makes it unworkable. If you are managing intake, log food and track weight over weeks, and treat the calorie burn figure as motivational decoration.
Use total sleep time, ignore the stage breakdown. The sleep versus wake number is well validated. The deep-versus-REM pie chart is a guess with roughly coin-flip reliability on some devices, and chasing it is how people develop anxiety about sleep, which is itself bad for sleep. Our sleep tracking guide goes further on reading these numbers without spiralling.
If you train by heart rate zones, wear a chest strap. This is the highest-value fix on the list, because zone training depends on accuracy exactly where wrist sensors are weakest.
The One Upgrade That Fixes the Biggest Gap
A chest strap measures the heart’s electrical signal directly, the same class of signal an ECG reads, rather than inferring pulse from light bouncing off a moving wrist. That is why the Polar H10 shows up as the criterion device in published research, including studies validating other wearables. When a paper needs a trustworthy heart rate reference outside a lab, this is frequently what they strap on.
If you will not wear a chest strap, and plenty of people genuinely will not, an armband is the honest compromise. The Polar Verity Sense uses optical sensing like your watch but reads from the upper arm, which has more muscle, less tendon movement, and far less of the interference that degrades wrist readings. It is not equal to a chest strap and it is clearly better than a wrist.
Check Polar Verity Sense Price →
For sleep specifically, the finger is a better optical site than the wrist for the same reason, which is part of why the Oura Ring 4 performs as it does in staging comparisons. Our Oura Ring 4 review and heart rate monitor roundup cover the buying side in full, and our true cost breakdown covers what these all cost over three years once subscriptions are counted.
FAQ
How accurate are fitness tracker heart rate monitors?
Good at rest and steady effort. In the Stanford validation, six of seven devices measured heart rate with under 5 percent median error. Accuracy declines as exercise intensity rises, and a systematic review of 32 studies found error as high as 20 percent depending on the model.
Are fitness tracker calorie counts accurate?
No. In the Stanford study, no device met the 5 percent accuracy threshold the researchers set. The most accurate was off by 27.4 percent and the least accurate by 92.6 percent. Use the figure for relative comparison at best, and never eat back exercise calories based on it.
Can a smartwatch accurately track sleep stages?
It tracks total sleep well and stages poorly. Sleep versus wake sensitivity was 95 percent or higher across devices in the 2024 Sensors validation, while stage sensitivity ranged from 50 to 86 percent. Treat the stage breakdown as an estimate rather than a measurement.
Why does my tracker say I slept when I was awake?
Because specificity for detecting wake is low. A 2025 six-device validation found specificity of just 29 to 52 percent, meaning devices over-label sleep. Lying still while awake commonly gets scored as sleeping.
Does skin tone affect fitness tracker accuracy?
The Stanford researchers reported that skin tone and body mass index both affected measurement accuracy, with darker skin and higher BMI associated with greater error. Optical sensors depend on light passing through skin, so this is a known limitation and remains under-researched.
What is the most accurate way to measure heart rate during exercise?
A chest strap, which reads the heart’s electrical activity directly rather than inferring it optically. Chest straps such as the Polar H10 are frequently used as the reference standard in validation research. An upper-arm optical band is the next best option.