What’s the Most Accurate Wearable? 45+ Studies, 7 Brands, Ranked (2026)

Updated: 23 hours ago
Last Updated: November 16, 2026

The most accurate wearable depends on what you are tracking, so the honest answer is one podium per metric. We ranked Oura Ring, Apple Watch, WHOOP, Garmin, Fitbit and Samsung Galaxy Watch on every peer-reviewed validation study we could find against a lab reference (polysomnography, chest-strap ECG, arterial blood, gas analysis), taking every figure from the paper’s own results tables. Oura wins everything measured overnight at rest: sleep stages, sleep timing, HRV and resting heart rate. Apple Watch wins heart rate during exercise and blood oxygen. Garmin wins steps and VO2 max. Fitbit is the least bad for calories, where no brand came within 9% on the max treadmill test. Skin temperature cannot be ranked because no brand has an independent sensor test. The podium is below, then the numbers behind every call. If your question is simply are fitness trackers accurate, our 71-study summary answers it in a few minutes.
We also built a free interactive comparison tool based on this research that lets you pick your devices and the metrics you care about to see them side by side: wearable accuracy comparison tool.
Most Accurate Wearable: The Podium
One podium per metric, ranked on the numbers across every measure of that metric, with sleep split into two because telling stages apart and knowing when you were asleep are different jobs with different winners. Brand-funded studies count and are flagged: the ones we rely on used the same method as the independent studies and agree with them. Research algorithms run on a watch’s raw data and unpublished company tests do not count. Each row uses the newest model with a published figure, so these are rankings of the models tested, mostly one to three generations old.
Metric | 1st | 2nd | 3rd | Confidence |
Sleep stages | Oura Ring (κ 0.65) | Apple Watch (κ 0.60) | Fitbit (κ 0.55) | High for 1st, medium after |
Sleep timing | Oura Ring (total sleep -2 min) | Fitbit (-3 min) | Apple Watch (+8 min) | High for 1st, medium after |
Overnight HRV | Oura Ring (6% error, agreement 0.99) | WHOOP (8%, 0.94) | Garmin (10.5%, 0.87) | High |
Resting heart rate | Oura Ring (1.94% error, 1.08 bpm) | Polar (2.71%, first 4 hours of sleep) | WHOOP (3.00%, 1.78 bpm) | Medium |
Exercise heart rate | Apple Watch (2.5 bpm cycling) | Garmin (2.9 bpm) | Fitbit (3.9 bpm cycling; 5.5% ten-device test) | Medium; top four within 2 to 7% |
Blood oxygen (SpO2) | Apple Watch (Arms 2.2) | Samsung (Arms 3.8) | Garmin (3.5 vs arterial blood) | High for 1st, medium after |
Steps | Garmin (+3% in daily life, 59 studies pooled) | Apple Watch (+7%) | Samsung (10.5%) | Medium |
Calories | Fitbit (14% on a max treadmill test) | Samsung (9% treadmill, 85% weights) | Apple Watch (26%) | Medium; no brand within 9% on the max treadmill test |
VO2 max | Garmin (6.7%, chest strap) | Fitbit (8.7%) | Apple Watch (13.2%) | High for 1st, medium after |
Breathing rate | Fitbit (0.3 to 1 breath/min high) | WHOOP (+0.1 breath/min) | No qualifying third | Low |
Skin temperature | Not ranked | Not ranked | Not ranked | No independent sensor test exists |
Models tested: sleep is Oura Ring 3, Apple Watch Series 8 and Fitbit Sense 2; HRV is Oura Ring 4, WHOOP 4.0 and Garmin Fenix 6; resting heart rate is Oura Ring 4, Polar Grit X Pro and WHOOP 4.0; exercise heart rate is Apple Series 8, Garmin Forerunner 955 and Fitbit Sense 2 cycling, with the Charge 6 and Vivoactive 5 in the ten-device test; SpO2 is Apple Series 8, Galaxy Watch 5 and Garmin Fenix 6 Pro; VO2 max is Garmin Forerunner 245, Pixel Watch 3 and Apple Series 10; breathing rate is Fitbit Charge 4 and WHOOP Strap 2.0. Each metric section names the model beside every figure.
Count the podium and the pattern is plain. Oura takes four firsts and places nowhere else: it was never tested for blood oxygen or VO2 max, its one lab calorie test (Oura Ring Gen 2, an older model) is context and not ranked, and it was 15% off on exercise heart rate and 50 to 69% off on steps. Apple Watch has two firsts, two seconds and three thirds. Garmin and Fitbit have two firsts each. WHOOP places only overnight because it has no peer-reviewed test for anything in the daytime. Samsung is second for blood oxygen and calories and third for steps. Polar places once, second for resting heart rate.
Sleep Stages: Oura, Apple, Fitbit
Can the device tell light, deep and REM apart. Kappa measures agreement with a lab sleep study (polysomnography): 1 is perfect, and two human scorers agree at about 0.75. Healthy-adult figures come from Robbins 2024 (Oura-funded, one lab test of all three leaders, same method as the independent studies), Svensson 2024 (Oura-funded; Oura deep and REM, 96 adults over several nights) and Miller 2022 (WHOOP 3.0 and Garmin Forerunner 245). The sleep-problems column is Schyvens 2025 (about half with sleep apnea), Lee 2023 (Oura, Samsung) and Herberger 2025 (sleep-clinic patients).
Rank | Brand | Model | Kappa, healthy | Deep sleep caught, healthy | REM caught, healthy | Kappa, sleep problems |
1 | Oura Ring | Ring 3 | 0.65 | 64% | 73% | 0.35 |
2 | Apple Watch | Series 8 | 0.60 | 51% | 83% | 0.53 |
3 | Fitbit | Sense 2 (Charge 5 clinical) | 0.55 | 62% | 68% | 0.41 |
WHOOP | 3.0 (4.0 clinical) | 0.44 | 62% | 66% | 0.37 | |
Samsung | Galaxy Watch 5 | not tested | not tested | not tested | 0.42 | |
Garmin | Forerunner 245 (Vivosmart 4 clinical) | 0.25 | 28% | 50% | 0.21 |
Apple beats Fitbit on two of the three measures (overall kappa and REM) and loses on deep sleep, where it scores half of real deep sleep as light. The Oura funding is footnoted, not discounted: the only independent head-to-head of current models (Topalidis 2026, a preprint) orders Oura 0.59, WHOOP 0.54 and Garmin 0.41, the same order. In people with sleep problems every device drops, and Apple holds up best (0.53 against Oura’s 0.35).
Sleep Timing: Oura, Fitbit, Apple
Does the device know when you were asleep. This is the part that drives your sleep score and your total sleep time. Figures are device minus the lab study, so + means the device reads high. Healthy adults: Svensson 2024 (Oura), Robbins 2024 (Apple, Fitbit), Miller 2022 (WHOOP 3.0, Garmin Forerunner 245), Chinoy 2021 (Garmin Vivosmart 3 efficiency, fall asleep and awake at night), Miller 2021 and Stone 2020 (WHOOP 2.0 fall asleep and efficiency).
Rank | Brand | Model | Awake time caught | Total sleep | Sleep efficiency | Time to fall asleep | Awake at night |
1 | Oura Ring | Ring 3 | 73% | -2 min | -1.1 pts | 0 min | -2 min |
2 | Fitbit | Sense 2 | 68% | -3 min | -1 pt | 0 min | +4 min |
3 | Apple Watch | Series 8 | 52% | +8 min | +2 pts | +2 min | -10 min |
WHOOP | 3.0 (2.0 for efficiency, fall asleep) | 56% | -12 min | +7.4 pts (vs an EEG headband) | 0 min | not reported | |
Garmin | Forerunner 245 (Vivosmart 3 for efficiency, fall asleep, awake at night) | 27% | +44 min | +10.1 pts | -1 min | -48 min | |
Samsung | none | no healthy-adult test |
Fitbit beats Apple on five of six timing measures. Apple’s one win is asleep-vs-awake agreement (93% against 91%), which runs high for every device because most of the night is sleep. Apple misses about half of the time you lie awake, so it inflates total sleep and reads time awake at night 10 minutes low. In people with sleep problems every brand reads total sleep 11 to 25 minutes high (Garmin Forerunner 945: 83 minutes) and none catches more than 52% of wake (Schyvens 2025, Herberger 2025, Budig 2022).
The full sleep breakdown, including sleep/wake agreement and manufacturer claims, is in our most accurate sleep tracker comparison and the sleep tracker accuracy tool.
Nocturnal HRV (Heart Rate Variability) Accuracy
An Ohio State University / Air Force Research Lab study (Dial et al., 2025) validated nocturnal HRV across 13 participants and 536 nights using a Polar H10 ECG chest strap as reference. No industry funding disclosed.
Rank | Brand | Model | Error (MAPE) | Agreement (CCC) | Reads high or low | Nights |
1 | Oura Ring | Ring 4 | 5.96% | 0.99 | -0.96 ms | 139 |
2 | WHOOP | 4.0 | 8.17% | 0.94 | -0.78 ms | 289 |
3 | Garmin | Fenix 6 | 10.52% | 0.87 | -1.84 ms | 150 |
Polar | Grit X Pro (first 4 hours of sleep) | 16.32% | 0.82 | not reported | 206 | |
Apple Watch | Series 9 / Ultra 2 (SDNN, mornings) | 28.9% | not reported | -8.31 ms | 316 readings | |
Fitbit | Charge 4 | not reported | ICC 0.69 | -7 ms | 7 days | |
Samsung | none | no test of the watch feature |
Oura Gen 3 scored 7.15% and CCC 0.97 in the same study, so the Ring 4 result is not a one-off. Apple reports SDNN rather than RMSSD and its 29% comes from morning readings (O’Grady 2024), so it is not directly comparable; Fitbit’s figures are from Hermans 2024. The Garmin Fenix 6 is two generations behind current hardware and the study authors said so. For what each brand's HRV number actually measures, and how the ranking holds up in other studies, see our full breakdown of the most accurate HRV wearable.
Resting Heart Rate Accuracy
From the same Ohio State study (Dial et al., 2025):
Rank | Brand | Model | Error (MAPE) | Typical error | Reads high or low |
1 | Oura Ring | Ring 4 | 1.94% | 1.08 bpm | -0.94 bpm |
2 | Polar | Grit X Pro (first 4 hours of sleep) | 2.71% | not reported | not reported |
3 | WHOOP | 4.0 | 3.00% | 1.78 bpm | -1.41 bpm |
Apple Watch | Series 9 / Ultra 2 (mornings, lying down) | 5.91% | 3.73 bpm | +0.5 bpm | |
Fitbit | Charge 4 | not reported | 1.16 bpm (Huntington’s patients) | +4 bpm in healthy adults | |
Garmin | Forerunner 245 (one lab night) | not reported | 5.4 bpm | +5.0 bpm | |
Samsung | none | no test of the watch feature |
Polar is second on 2.71%, a figure that covers the first four hours of sleep. Polar and WHOOP are 0.3 points apart, within noise. Fitbit reads about 4 bpm high in healthy adults because its resting figure includes relaxed awake time (Hermans 2024). The Garmin Fenix 6 was excluded from the Dial resting heart rate analysis over timestamp alignment; its 5.4 bpm is the Forerunner 245 in one lab night (Miller 2022). Sources: Dial 2025; O’Grady 2024; Doheny 2024; Hermans 2024; Miller 2022.
Daytime and Exercise Heart Rate Accuracy
Correction, September 2026: an earlier version of this section used WellnessPulse figures (Apple Watch 86.3%, Fitbit 73.6%, Garmin 67.7%). WellnessPulse is a consumer aggregator, and its accuracy percentages are pooled correlation coefficients expressed as percentages, not accuracy rates. They have been replaced with peer-reviewed data. The largest single-protocol test is Gielen et al. (2026, KU Leuven, no conflicts): ten devices, one Zephyr chest strap reference, 45 participants, seated rest, a stress task, treadmill walking and intermittent walking. Only two watches were worn per session, so the range is quotable and the exact order is not. Read it as tiers.
Tier | Device | Median MAPE | Agreement (CCC) |
1 | Fitbit Charge 6 | 5.5% | 0.93 |
1 | Garmin Vivoactive 5 | 6.3% | 0.83 |
1 | Google Pixel Watch 2 | 6.7% | 0.87 |
2 | Apple Watch SE | 7.3% | 0.70 |
2 | Garmin Vivosmart 5 | 8.1% | 0.78 |
3 | Polar Ignite 3 | 11.2% | 0.63 |
3 | Xiaomi Watch 2 | 11.9% | 0.69 |
3 | Polar Pacer | 13.1% | 0.66 |
4 | Oura Ring Gen 3 | 15.0% | 0.61 |
4 | Fitbit Inspire 3 | 16.5% | 0.45 |
Three head-to-head tests put the brands on the same people, and together they set the podium: Apple first, Garmin second, Fitbit third, Samsung fourth. The ten-device test above is one of them; the other two are a cycling test against a chest strap (Kostrna 2026) and a treadmill test against a 12-lead ECG (Van Oost 2025).
Rank | Brand | Cycling, 4 watches (MAE) | Ten-device test (MAPE) | Treadmill vs 12-lead ECG (MAPE) | Other |
1 | Apple Watch | Series 8: 2.51 bpm (best of 4) | SE: 7.3% | not tested | Series 8 bias 0.0 in a second cycling test; pooled bias -0.27 bpm across 22 studies |
2 | Garmin | Forerunner 955: 2.93 bpm | Vivoactive 5: 6.3% | Vivosmart 4: 4.40%, with 50% of readings missing | Vivosmart 5: 20% of readings missing |
3 | Fitbit | Sense 2: 3.89 bpm (last of 4) | Charge 6: 5.5%; Inspire 3: 16.5% | Sense 2: 5.61%; Charge 5: 5.65% | Pixel Watch 3: 2.73% on a max treadmill test (Google-funded) |
4 | Samsung | Galaxy Watch 5: 3.38 bpm | not tested | not tested | Galaxy Watch 4 under 2% in heart patients on a stress test; Galaxy Watch 6 worst of 4 on a treadmill |
WHOOP | not tested | not tested | 4.0: 8.52% | ||
Oura Ring | not tested | Gen 3: 15.0% | not tested |
The top four are within 2 to 7% of each other, so the order is not firm. Samsung’s often-quoted sub-2% result came from heart patients on a hospital stress test (Kim 2023); in the head-to-heads it was third of four cycling and worst of four on a treadmill (Jeong 2026). Sources: Kostrna 2026; Gielen 2026; Van Oost 2025; Mulholland 2025; Lambe 2026.
The failure mode is irregular arm movement, not intensity. Steady running is one of the easiest conditions for a wrist device; badminton, soccer and rowing are the worst. Night is easy and day is hard, which is why every manufacturer claim (WHOOP 99.7%, Oura 99%) comes from sleep data. Two free fixes with measured effect sizes: wear the watch three finger-widths above the wrist bone and tighten the strap.
Full device tiers and activity-by-activity data are in the heart rate accuracy tool, the ten-device heart rate ranking and heart rate accuracy by activity type.
Blood Oxygen (SpO2) Accuracy
Arms is the FDA and ISO measure, rated over 70 to 100% saturation; 4.0 or less passes. None of these SpO2 features are FDA-cleared; they are wellness features.
Rank | Brand | Model | Arms, low-oxygen lab test | Bias, low-oxygen test | Error vs arterial blood, ward patients | Other |
1 | Apple Watch | Series 8 (Series 7 vs arterial blood) | 2.2 | -0.1 | 2.44 (112 patients) | Pooled bias -0.04 across 9 studies; Arms 4.6 in 9 people taken to 56% saturation, outside the rated range |
2 | Samsung | Galaxy Watch 5 (Watch 4 vs arterial blood) | 3.8 | -2.6 (reads low) | 1.8 under the FDA protocol (Samsung-funded) | 2.5 vs sleep-lab oximetry (Samsung-funded) |
3 | Garmin | Fenix 6 Pro (Venu 2s at rest) | not tested | not tested | 3.50, reads 2.7 low (109 patients) | 4.10 vs a finger oximeter (fails the ISO limit); Venu 2s read 5.5 low in lung patients; Fenix 6 barely moved when oxygen fell |
Fitbit | Charge 4, Sense 2 | not tested | not tested | not tested | +2, ICC 0.32 (COPD, overnight); +1.7, ICC 0.47 (inpatients) | |
Oura Ring | none | no peer-reviewed test | ||||
WHOOP | none | no peer-reviewed test |
Apple and Samsung come from the same induced-hypoxia study (Walzel 2023). Every watch drifts when oxygen is truly low and fails to read 10 to 45% of the time; an Apple Watch Series 7 over-read 85% of readings below 88% against arterial blood (Jiang 2026). Sources: Walzel 2023; Helmer 2023, 2024; Jiang 2023, 2026; Browne 2024, 2025; Carrier 2025; Hermans 2024; Lee 2026; Lambe 2026.
Step Count Accuracy
Correction, September 2026: the earlier table here (Garmin 82.6%, Apple 81.1%, Fitbit 77.3%) was WellnessPulse data, which are pooled correlations rather than accuracy rates. Replaced with peer-reviewed error rates. Lab and free-living are different tests, so both are shown.
Rank | Brand | Free living, pooled across 59 studies | Free living, single study | Lab walking |
1 | Garmin | +3% | Vivofit Jr 2: 9.6% (children) | Vivoactive 3, Vivofit 3: 1 to 2% in 258 adults; Vivosmart 4: 2.4%; Vivoactive 4: 4.6% |
2 | Apple Watch | +7% | Series 6: 6.4% (36 adults); +2.1% (21 adults) | Within 3.4% at every treadmill speed (model not stated); Series 5: 9.3% on slow and shuffle walks |
3 | Samsung | not pooled | Galaxy Watch 4: 10.5% | Galaxy Watch Active 2: -7.2% on slow and shuffle walks |
Fitbit | +15% | Inspire 2: 25.9%; Charge 4: 26.0%; Sense: +18% | Inspire 2: 3.6% at normal pace | |
Oura Ring | not pooled | Gen 2: 50 to 69% off, 2,000 to 3,800 steps a day too many; Gen 3: -6.2% (5 people) | no lab test | |
WHOOP | no published validation |
Fitbit is accurate on a treadmill but overcounts in daily life, which is how people use it, so it is fourth. All Oura figures predate its May 2025 step algorithm and nothing has been published since. Sources: Paluch 2026; Mora-Gonzalez 2022; Bianchini 2022; de Leon 2026; Mayorga-Vega 2023; Hong 2024; Miwa 2026; Veerabhadrappa 2018; Rowe 2025; Delobelle 2024; Waddell 2024; Cheung 2025; Kristiansson 2023; Henriksen 2022.
Full breakdown in the step count accuracy tool and which wearable has the most accurate step count.
Energy Expenditure (Calories) Accuracy
Calories are the least accurate number on any wearable, so this podium separates bad from worse. Four head-to-head studies put the same four brands on the same people against lab gas analysis; each column is one study. Error: lower is better. Reads high: closer to 0 is better.
Rank | Brand | Models | Max treadmill test (error) | Cycling (error) | Weight training (error) | Endurance (reads high) | Weight training (reads high) |
1 | Fitbit | Charge 6; Sense 2; Charge 5 | 14.0% | 31.7% | 41.7% | +21% | +4% |
2 | Samsung | Galaxy Watch 7; Galaxy Watch 5 | 9.4% | 66.5% | 84.5% | +9% | +83% |
3 | Apple Watch | Ultra 2; Series 8 | 26.0% | 35.7% | 112.2% | +14% | +112% |
4 | Garmin | Forerunner 265S; Forerunner 955; Vivosmart 5 | 14.7% | 77.6% | 114.9% | +30.5% | +116.4% |
Oura Ring | none | one lab test against gas analysis: 21.1% error for Oura Ring Gen 2 (older model), measured as METs, context only and not ranked | |||||
WHOOP | none | no calorie study |
Fitbit wins three of the five comparisons and has the smallest worst case; its cycling figure was computed after seven implausible readings were removed. Samsung is the most accurate on steady running and walking (10 to 13% running in a Samsung-funded test, 0.5% walking) but roughly doubles true burn on weights. The daily-total figures in earlier versions of this post (Oura 13%, Apple about 28%) used a different reference for each brand and cannot be compared, so they no longer set a rank. Sources: Perez 2025 (conference abstract, about 20 adults); Kostrna 2026 (58 adults); Yun 2026 (70 men); Lee 2026 (62 men); Ferreira 2026; Oh 2024.
Treat any calorie count as a trend, not a target. More detail: calorie burn accuracy calculator.
See our full breakdown of how accurate wearable calorie burn is by device.
VO2 Max Estimation Accuracy
Method beats brand. Estimates built from a real outdoor run have about zero group-level bias against the lab, while estimates built from resting heart rate overestimate by about 2 mL/kg/min (INTERLIVE meta-analysis, 2022). Either way individual error is wide. Accuracy degrades in highly trained users, and the one elite study that avoided this used a chest strap. See the full VO2 max accuracy comparison for how each device estimates it.
Rank | Brand | Model | Error (MAPE) | Bias (mL/kg/min) | Sample | Source |
1 | Garmin | Forerunner 245 (chest strap) | 6.7% | -4.05 | 35 runners and triathletes | Engel 2026 |
1 | Garmin | Fenix 6 (chest strap) | 7.05% | +0.75 | 19 adults | Carrier 2025 |
2 | Fitbit | Pixel Watch 3 | 8.65% | -0.53 | 26 adults | Lambe 2026 (Google-funded) |
2 | Fitbit | Charge 2 (retired algorithm) | 9.1% | +0.30 | 60 adults | Klepin 2019 |
3 | Apple Watch | Series 10 | 13.2% | -6.25 | 35 adults | Lambe 2026 |
3 | Apple Watch | Series 9 / Ultra 2 | 13.3% | -6.07 | 28 adults | Lambe 2025 |
3 | Apple Watch | Series 7 | 15.8% | -4.51 | 19 adults | Caserman 2024 |
Polar | Vantage M2 (resting test) | 13.7% | +1.0 | 24 adults, mostly patients | Neudorfer 2025 | |
Garmin | Vivoactive 4S, wrist only | 56% (median) | +13.4 | 207 mostly older women | Jamieson 2024 | |
Samsung, WHOOP, Oura Ring | none | no peer-reviewed study | WHOOP’s 8.0% is an unpublished internal test | none |
Garmin’s first place holds for runners with a chest strap; wrist-only in older adults it read 13 mL/kg/min high. Fitbit’s second place rests on the Pixel Watch 3, which uses the outdoor-run method Google ships today; the Charge 2 studies tested an algorithm Google retired in May 2026. Apple underestimates in all three independent studies, most in fit women.
For a device-by-device breakdown, see which wearable is most accurate for VO2 max, and what actually affects your VO2 max in the first place.
Breathing Rate: Fitbit, WHOOP, No Third
Overnight breaths per minute off a reference; within 3 is the usual clinical limit. Only two brands have a qualifying test of the feature a buyer gets.
Rank | Brand | Model | Independent test | Company test |
1 | Fitbit | Charge 4 | +0.3 to +1.0 breaths/min, ICC 0.77 to 0.84 (Hermans 2024) | RMSE 0.65 (Natarajan 2021, Fitbit staff) |
2 | WHOOP | Strap 2.0 | not tested | +0.1 breaths/min, precision 1.0, vs sleep-lab belts (Berryhill 2020, WHOOP-funded, peer reviewed) |
Garmin | Fenix 6 Pro; Venu 3 | -1.46, r 0.28 in ward patients (Helmer 2024); MAE 4.21 (Knevels 2026, preprint) | not tested | |
Apple Watch | Series 7 | No readings in two hospital studies | not tested | |
Oura Ring | not named | not tested | 0.71 (2020 blog post, not peer reviewed) | |
Samsung | Galaxy Watch 4 / 5 sensors | not tested | RMSE 1.13 for a research algorithm, not the watch feature (Jung 2023, Samsung-funded) |
Garmin’s shipped feature failed its hospital test, Apple only measures during sleep and gave no readings, and the Oura and Samsung figures do not qualify. Low confidence: the Fitbit and WHOOP figures use different references and WHOOP’s is a 2020 strap.
Skin Temperature Accuracy
Oura's own validation, a company blog post from December 2020 that was not peer reviewed, tested its rings two ways:
In the lab, rings tested on their own rather than worn: r² > 0.99 against research iButton sensors, changes as small as 0.13°C detected, accurate to within 0.36°C. Worn by 16 people for one week: r² > 0.92, with no error figure reported.
⚠️ Funding: This is Oura's own test, not peer reviewed, and r² measures how closely the ring tracks a reference rather than how far off it reads. The ovulation study sometimes cited as validation (Maijala et al., 2019) was co-authored by Oura employees. No independent study has tested any consumer wearable's temperature sensor against a reference thermometer, as we cover in wearable skin temperature accuracy.
FDA-Cleared Features
Most wearable metrics are wellness estimates. A few features have earned FDA authorization:
Device | Feature | Status |
Apple Watch (Series 4+) | ECG / Atrial Fibrillation Detection | FDA Cleared |
Samsung Galaxy Watch (4+) | ECG / Atrial Fibrillation Detection | FDA Cleared |
Apple Watch (Series 9+, Ultra 2) | Sleep Apnea Notification | FDA Authorized |
Samsung Galaxy Watch | Sleep Apnea Detection | FDA Authorized (Feb 2024) |
Apple Watch | Blood Oxygen (SpO2) | Wellness feature (NOT FDA cleared) |
Fitbit | Irregular Rhythm Notification | FDA Cleared |
Important Caveats
Before drawing conclusions from any of this data, keep these limitations in mind:
No single device wins everywhere. The best device depends on which metric matters most to you.
Healthy vs clinical is the biggest gap. The same device scores far higher in screened healthy young sleepers than in people with sleep complaints: Oura Gen 3 goes from κ=0.65 to 0.35, Apple from 0.60 (Robbins 2024, healthy) to 0.30 (Lee 2023, clinical). Numbers from healthy cohorts do not describe most people reading this.
Study funding matters. The primary sleep study (Robbins et al.) was Oura-funded. Independent studies (Schyvens, Lee, Herberger) found different rankings. The funded study used the same method; its higher number comes from its screened population.
Current hardware is barely studied. Peer review lags hardware by 2 to 4 years. Oura Ring 5, WHOOP 5.0 and MG, Apple Watch Series 10, 11 and Ultra 3, Galaxy Watch 7 and 8, Pixel Watch 3 and 4, and every Garmin flagship since the Fenix 6 have no independent validation for any metric here. Validation of one generation does not transfer to the next.
Small sample sizes. The HRV/RHR study had 13 participants (536 nights). Antwerp had 62 participants, 1 night each. Brigham had 35 participants over multiple nights. The ten-device heart rate test had 10 sessions per device.
Mean bias is the wrong metric. A bias near zero means overestimates and underestimates cancelled. Read the error rate and the limits of agreement.
All wearables are estimates. None are medical devices (except specific FDA-cleared features listed above). Data should inform, not diagnose.
Individual variation. Accuracy varies with skin tone, tattoos, BMI, wrist fit, and activity level. Most validation studies have predominantly Caucasian participants, a documented research gap.
PSG is imperfect too. The gold standard polysomnography has interrater reliability of about κ=0.75, meaning even human experts disagree about a quarter of the time on sleep staging.
Common device failure mode. All consumer devices tend to misclassify wake, deep sleep, and REM as light sleep, a conservative approach that inflates light sleep totals.
Why Accuracy Matters for Understanding Food-Biometric Patterns
If you’re trying to understand how nutrition affects your sleep, recovery, or energy levels, the accuracy of your wearable data is the foundation everything else builds on. When measurement error is high, real patterns between what you eat and how your body responds get harder to detect. When accuracy is high, the data can surface connections, like how meal timing affects your overnight heart rate, or whether a supplement is actually changing your HRV, that you’d never spot manually.
This is part of the reason we built Kygo Health to integrate with multiple wearable platforms. Different devices bring different strengths. Connecting them to nutrition data in one place gives you a more complete picture to work with.
Using Multiple Wearables Together
Many people in the biohacking and quantified self communities wear multiple devices simultaneously to capture different metrics from different strengths, Oura Ring for sleep plus Apple Watch for workouts, or WHOOP plus Garmin for different contexts.
The challenge is getting that data to talk to each other. We wrote a detailed guide on this: How to Centralize Health Data from Multiple Devices.
If you’re specifically using Oura for sleep and want to connect that with food tracking, check out: How to Combine Oura Ring with Food Tracking.
Want to compare devices yourself? Explore all the data from these studies in our free Wearable Accuracy Comparison Tool.
Now you know which device is most accurate for each metric. The harder question is what your own numbers actually respond to.
Sources
Robbins R, et al. (2024). Accuracy of Three Commercial Wearable Devices for Sleep Tracking in Healthy Adults. Sensors, 24(20), 6532. DOI: 10.3390/s24206532. Funded by Oura Ring Inc.
Schyvens AM, et al. (2025). A performance validation of six commercial wrist-worn wearable sleep-tracking devices for sleep stage scoring compared to polysomnography. Sleep Advances, 6(2), zpaf021. DOI: 10.1093/sleepadvances/zpaf021. Independent (VLAIO-funded).
Lee T, et al. (2023). Accuracy of 11 Wearable, Nearable, and Airable Consumer Sleep Trackers. JMIR mHealth and uHealth, 11, e50983. DOI: 10.2196/50983. Independent; clinical sample; several authors affiliated with Asleep Co.
Herberger S, et al. (2025). Oura Ring Gen 3 in sleep-clinic patients. Scientific Reports, 15, 9461. Independent.
Khan S, et al. (2025). Oura Ring sleep meta-analysis, 6 studies, N=388. OTO Open, 9, e70181. Independent.
Dial MB, et al. (2025). Validation of nocturnal resting heart rate and heart rate variability in consumer wearables. Physiological Reports, 13(16), e70527. DOI: 10.14814/phy2.70527. Air Force Research Laboratory funded.
Gielen W, et al. (2026). Ten-device wrist heart rate validation. JMIR Formative Research, e85186. KU Leuven, no conflicts.
Van Oost L, et al. (2025). Five consumer wearables versus 12-lead ECG. Sensors, 25(20), 6319. VLAIO-funded.
Kim J, et al. (2023). Apple Watch 7 and Galaxy Watch 4 versus 12-lead ECG during cardiopulmonary exercise testing. Annals of Rehabilitation Medicine. Independent.
Shcherbina A, et al. (2017). Accuracy in wrist-worn, sensor-based measurements of heart rate and energy expenditure in a diverse cohort. Journal of Personalized Medicine, 7(2), 3.
Choe J, Kang M (2025). Apple Watch accuracy meta-analysis, 56 studies. Physiological Measurement.
Chevance G, et al. (2022). Accuracy and precision of energy expenditure, heart rate, and steps measured by Fitbit. JMIR mHealth and uHealth, 10(4), e35626.
Kristiansson E, et al. (2023). Validation of Oura ring energy expenditure and steps in laboratory and free-living. BMC Medical Research Methodology, with correction DOI: 10.1186/s12874-023-02029-w.
Niela-Vilén H, et al. (2022). Oura ring versus ActiGraph in free living. Computers, Informatics, Nursing, 40(12), 856-862.
Miwa M, et al. (2026). Apple Watch Series 6 and Fitbit Sense step counts over three weeks versus ActiGraph GT9X. PLOS ONE.
Kim, et al. (2024). Apple Watch Series 6 free-living step count versus ActivPAL. Sensors.
Rowe, et al. (2025). Apple Watch Series 5 step count at slow and shuffle walking speeds. PLOS ONE.
de Leon, et al. (2026). Garmin Vivoactive 4 treadmill step count and heart rate. Applied Sciences, 16(3), 1286.
Rider BC, et al. (2025). Four sports watches in laboratory and field. Journal for the Measurement of Physical Behaviour, 8(1).
Cheung YS, et al. (2025). Validity of Fitbit Inspire 2 in step count during treadmill walking. Physiotherapy Practice and Research, 46(2), 95-102.
Parak J, et al. (2017). Energy expenditure estimation with a wrist device using Firstbeat modeling. JMIR mHealth and uHealth.
Molina-García P, et al. (2022). Validity of estimating VO2 max by consumer wearables: INTERLIVE systematic review and meta-analysis. Sports Medicine, 52(7).
Carrier B, et al. (2025). Validation of aerobic capacity and pulse oximetry in wearable technology. Sensors, 25(1), 275.
Carrier B, et al. (2023). Validation of aerobic capacity and lactate threshold in wearable technology for athletic populations. Technologies, 11(3), 71.
Engel FA, et al. (2026). Validity of VO2max estimates from the Forerunner 245 in highly vs moderately trained endurance athletes. European Journal of Applied Physiology, 126, 591-603.
Caserman P, et al. (2024). Assessing the accuracy of smartwatch-based estimation of maximum oxygen uptake using the Apple Watch Series 7. JMIR Biomedical Engineering, 9, e59459.
Lambe R, et al. (2025). Investigating the accuracy of Apple Watch VO2 max measurements. PLOS ONE, 20(5), e0323741.
Apple Watch Series 10 VO2 max validation (2026). Mayo Clinic Proceedings: Digital Health.
Freeberg KA, et al. (2019). Assessing the ability of the Fitbit Charge 2 to accurately predict VO2max. mHealth, 5, 39. Klepin K, et al. (2019). Medicine and Science in Sports and Exercise, 51(11), 2251-2256.
Neudorfer, et al. (2025). Polar VO2 max estimate versus cardiopulmonary exercise testing.
Christakis, et al. (2025). A guide to consumer-grade wearables in cardiovascular clinical care. npj Cardiovascular Health, 2, 82.
Kryder C. (2020). How Accurate Is Oura's Temperature Data? Oura company blog, 7 December 2020. Not peer reviewed.
Maijala A, et al. (2019). Nocturnal finger skin temperature in menstrual cycle tracking. BMC Women's Health, 19, 150. Two of five authors employed by Oura.
Lanfranchi, et al. (2024). Samsung Galaxy Watch SpO2 validation. Journal of Clinical Sleep Medicine.
Svensson T, et al. (2024). Validity of the Oura Ring Gen 3 against polysomnography over multiple nights in healthy adults. Sleep Medicine. DOI: 10.1016/j.sleep.2024.01.020. Oura-funded authors.
Miller DJ, et al. (2022). A validation of six wearable devices for estimating sleep, heart rate and heart rate variability in healthy adults. Sensors, 22(16), 6317. DOI: 10.3390/s22166317. Lab receives WHOOP support.
Topalidis P, et al. (2026). Head-to-head comparison of eight consumer sleep trackers against polysomnography. Preprint, not peer reviewed.
Chinoy ED, et al. (2021). Performance of seven consumer sleep-tracking devices compared with polysomnography. Sleep, 44(5). DOI: 10.1093/sleep/zsaa291.
Budig M, et al. (2022). Validity of two consumer multisport activity trackers and one accelerometer against polysomnography. Sensors, 22(23), 9540. DOI: 10.3390/s22239540.
Kostrna J, et al. (2026). Heart rate and energy expenditure accuracy of four smartwatches during cycling in Hispanic adults. PLOS ONE. DOI: 10.1371/journal.pone.0353261.
Yun S, et al. (2026). Energy expenditure accuracy of four wearables during resistance training. Bioengineering, 13(10), 1099. DOI: 10.3390/bioengineering13101099.
Lee, et al. (2026). Accuracy of four wearables for heart rate and energy expenditure in endurance and resistance exercise, 62 men. Sensors. SOLUM Health Care grant; one author employed.
Perez, et al. (2025). Energy expenditure accuracy of five wearables during a maximal treadmill test. International Journal of Exercise Science: Conference Proceedings, 14(5), 102. Conference abstract.
Ferreira, et al. (2026). Galaxy Watch 6 and 7 energy expenditure during intermittent running. JMIR Formative Research. DOI: 10.2196/83090. Samsung-funded.
Walzel S, et al. (2023). Blood oxygen accuracy of the Apple Watch Series 8, Galaxy Watch 5 and Withings ScanWatch under induced hypoxia. Sensors.
Helmer P, et al. (2023). Smartwatch pulse oximetry against arterial blood gas in post-surgical patients. iScience. Helmer P, et al. (2024). Vital-sign accuracy of consumer smartwatches on a surgical ward. Digital Health. DOI: 10.1177/20552076241254026.
Paluch AE, et al. (2026). Free-living step count accuracy of wrist wearables against a hip accelerometer: a network meta-analysis of 59 studies. Journal of Sport and Health Science.
Hong KR, et al. (2024). Apple Watch Series 6 and Galaxy Watch 4 step counts against activPAL over 24 hours. Sensors, 24(14), 4658. DOI: 10.3390/s24144658.
Lambe R, et al. (2026). Validity of the Google Pixel Watch 3 for VO2 max and heart rate against a maximal treadmill test. PLOS ONE. DOI: 10.1371/journal.pone.0356808. Google-funded.
Lambe R, et al. (2026). The accuracy of Apple Watch measurements: a living systematic review and meta-analysis. npj Digital Medicine. DOI: 10.1038/s41746-025-02238-1.
Mulholland, et al. (2025). Apple Watch Series 8 and Garmin Vivosmart 5 heart rate during cycling across skin tones. European Journal of Applied Physiology.
Jeong, et al. (2026). Heart rate accuracy of four wearables on a treadmill. Physical Activity and Nutrition. Abstract.
O’Grady B, et al. (2024). Apple Watch Series 9 and Ultra 2 HRV and resting heart rate against a chest strap over 14 days. Sensors.
Hermans L, et al. (2024). Fitbit Charge 4 heart rate, HRV, SpO2 and breathing rate in COPD and healthy adults. JMIR mHealth and uHealth. DOI: 10.2196/56027.
Natarajan A, et al. (2021). Measurement of respiratory rate using wearable devices and applications to COVID-19 detection. npj Digital Medicine. DOI: 10.1038/s41746-021-00493-6. All authors Fitbit staff.
Berryhill S, et al. (2020). Effect of wearables on sleep in healthy individuals: a randomized crossover trial and validation study. Journal of Clinical Sleep Medicine. DOI: 10.5664/jcsm.8356. WHOOP-funded.
Doheny EP, et al. (2024). Fitbit Charge 4 sleep and heart rate against polysomnography in Huntington’s disease. Journal of Clinical Sleep Medicine. DOI: 10.5664/jcsm.11098.
Mora-Gonzalez J, et al. (2022). Step count accuracy of Garmin wrist devices across treadmill speeds in 258 adults. International Journal of Behavioral Nutrition and Physical Activity.
Bianchini E, et al. (2022). Garmin Vivosmart 4 step count in Parkinson’s disease and healthy adults. Sensors.
Jamieson A, et al. (2024). Wrist-based VO2 max estimates from the Garmin Vivoactive 4S in 207 adults. Computing in Cardiology.
FAQ: Wearable Accuracy Questions
Which wearable is the most accurate overall?
No single wearable wins everywhere, so we rank one podium per metric. Oura Ring is the most accurate for everything measured overnight at rest: sleep stages, sleep timing, HRV and resting heart rate. Apple Watch is the most accurate for heart rate during exercise and blood oxygen, with the most FDA-cleared features. Garmin is the most accurate for steps and, with a chest strap, VO2 max. Fitbit is the least bad for calories and the only brand with an independent breathing rate test. WHOOP is second for HRV and third for resting heart rate, where Polar is second; Samsung is second for blood oxygen and calories.
Which wearable is the most accurate for sleep tracking?
Oura Ring, for healthy sleepers. In the one study that put Oura, Apple and Fitbit on the same sleepers (Robbins 2024, Oura-funded, same method as the independent studies), Oura scored κ 0.65, Apple 0.60 and Fitbit 0.55, and Oura also leads sleep timing with total sleep within 2 minutes of the lab. Apple catches REM best but misses about half of the time you lie awake. In people with sleep problems every device drops, and Apple holds up best (κ 0.53).
How accurate is Oura Ring HRV compared to medical devices?
Oura Ring 4 achieved a 0.99 concordance correlation coefficient with a Polar H10 ECG chest strap in an independent 536-night study (Dial et al., 2025), the highest HRV accuracy among the consumer wearables tested. That is nocturnal HRV only; Oura's daytime heart rate ran 15% error in a separate ten-device test. Oura Ring 5 has not been validated.
Is WHOOP accurate for HRV tracking?
WHOOP 4.0 showed a CCC of 0.94 and MAPE of 8.17% for nocturnal HRV in the Dial et al. (2025) study, rated Moderate on the concordance scale. WHOOP's 99.7% marketing figure comes from a one-night sleep study on the WHOOP 3.0 and does not describe daytime accuracy, where WHOOP 4.0 ran 8.5% error against a 12-lead ECG.
Does skin tone affect wearable accuracy?
PPG sensor accuracy is affected by skin pigmentation. Most validation studies have predominantly Caucasian participants, which is a known research gap. Accuracy data may not generalize equally across all skin tones.
Can I use multiple wearables together?
Yes. Many people wear multiple devices to capture different metrics from each device’s strengths. The challenge is correlating data across platforms, which typically requires a third-party platform or manual comparison.
Which wearable is best for tracking how food affects sleep?
For nutrition-sleep correlation analysis, you need accurate sleep and HRV data paired with consistent food logging. The studies above show which devices perform strongest for each metric. The best choice depends on which signal you care about most: a ring for nocturnal HRV and resting heart rate, a watch for sleep stage labels and daytime heart rate.
Are wearable calorie estimates reliable?
No: no brand came within 9% on the max treadmill test. Fitbit is the least bad (14% on a max treadmill test, 32% cycling, 42% on weights), Samsung is closest on steady running and walking, and Apple and Garmin are off by more than 110% on weight training. Treat calorie counts as a trend, not a number.
Why do different studies show different accuracy rankings?
Study funding, sample size, population demographics, device firmware version, number of nights tested, and PSG scoring protocols all affect results. This is why we include multiple studies and flag funding sources throughout this article.
Disclaimer: Kygo Health LLC is a personal data aggregation and insights platform designed for informational purposes only. The information provided does not constitute medical advice, diagnosis, or treatment. Always consult a licensed healthcare provider with any questions regarding medical conditions.
Have questions about wearable accuracy or data you think should be included? Reach out directly at Ryan@kygo.app. If you have sources or credible data that isn’t listed here, share it and we’ll review and update accordingly.