Which Wearable Has the Most Accurate Step Count? A 2024-2025 Research Analysis

Last Updated: March 22, 2026

This is everything we could find on step count accuracy across consumer wearables — 20+ peer-reviewed studies (2020–2026) plus consumer testing for devices that haven't been studied.

All 27 sources are listed at the bottom. We flagged study types and funding so you can judge for yourself. Research lags behind product releases, so some older models are included. Device generation matters — results from older hardware may not apply to current models.

For a breakdown of what factors affect your accuracy more than device choice — walking speed, arm swing, placement, age — see our companion post: What Actually Affects Your Step Count Accuracy.

To compare devices visually, use our free Step Count Accuracy Comparison Tool.

Step Count Accuracy: Master Summary

Device	MAPE (Lab)	MAPE (Free-Living)	Bias Direction	Peer-Reviewed Studies
🥇 Garmin	0.6–3.5%	10–17.8%	Underestimates	42
🥈 Apple Watch	0.9–3.4%	6.4–10%	Slight underestimate	28+
🥉 Fitbit	~5–8%	10–25%	Mixed	144
Samsung Galaxy Watch	Limited data	Limited data	Overestimates	1
Oura Ring	<10% (controlled)	~42% overcount	Overestimates	2
Polar	—	—	Overestimates (older)	1
COROS	—	—	Slight underestimate	0
WHOOP	—	—	Unknown	0
Google Pixel Watch	—	—	Overcounted (post-update)	0

Notes:

• Study counts from Fuller et al. (2020), JMIR

• "Overall accuracy" percentages circulating online (82.6%, 81.1%, 77.3%) come from WellnessPulse — a consumer data aggregation, not peer-reviewed research

• MAPE ranges above come from actual published studies

Garmin

Garmin has 42 peer-reviewed step validation studies — among the most of any consumer brand.

Peer-reviewed data:

Model	MAPE	Condition	Source
Vivoactive 4	<2%	Exercise testing	Nature Scientific Reports (2024)
Vivosmart HR	0.61–1.27%	Treadmill (3.2–4.8 km/h)	Feehan et al. (2020)
Vivosmart HR+	≤5% treadmill, ≤10% free motion	All conditions	Roos et al. (2020)
Vivosmart	1.2–3.5%	Three treadmill speeds	Garmin validity review (2020)
Vivofit	17.8%	Free-living (at home)	Garmin validity review (2020)

Garmin uses a 10-step bout threshold — it won't register steps until 10+ consecutive patterns are detected, then retroactively credits all 10. This reduces phantom steps but misses very short walks.

All Garmin devices consistently underestimate at slow speeds (<1.6 km/h). No peer-reviewed validation exists for any model newer than Vivoactive 4.

• Sources: Feehan et al. (2020); Roos et al. (2020); Garmin validity review (2020); Nature Scientific Reports (2024)

  
  

Consumer testing:

Model	Result	Condition	Source
Forerunner 265	0.3% error (15 steps off / 5,000)	Walking	Android Central (2023) ⚠️
Forerunner 970	Within 100 steps / 10,000	Walk + jog	Android Central (2025) ⚠️
Forerunner 970	~5,037 steps tracked from pocket	Pocket test	Android Central (2025) ⚠️

The FR970 pocket result suggests leg-motion detection without requiring wrist swing.

• Sources: Android Central (2023, 2025) ⚠️

Apple Watch

Apple Watch has 28+ peer-reviewed step validation studies.

Peer-reviewed data:

Model	MAPE	Condition	Source
Series 6	6.4%	Free-living (24h vs ActivPAL)	Kim et al. (2024)
Series 2	~18.5%	Mixed conditions	Choe & Kang (2025) meta-analysis

Series 6 achieved r=0.99 correlation vs ActivPAL in 24-hour free-living — the highest independent correlation among consumer wearables in that study. No peer-reviewed step validation exists for Series 8, 9, 10, or Ultra 2. A 2026 living review of 82 studies found no improvement trend across Apple Watch generations for step counting.

• Sources: Kim et al. (2024); Choe & Kang (2025); npj Digital Medicine (2026)

MAPE by activity type (Choe & Kang 2025 meta-analysis across models):

Activity	MAPE
Moderate-to-vigorous walking	3.0%
Non-treadmill walking	<10%
Treadmill	10.1%
Light-intensity / slow walking	23.9%

Slow walking MAPE (23.9%) is 8x worse than brisk walking (3.0%). Treadmill produces worse results than overground walking.

MAPE by age (Choe & Kang 2025):

Age Group	MAPE
Under 40	4.3%
40 and older	10.9%

A 2.5x accuracy gap exists between age groups with the same device.

• Sources: Choe & Kang (2025)

Consumer testing:

Model	Result	Condition
Ultra 2	Most consistent across all 3 tests	10-watch walk/jog comparison
Ultra 2	~5,088 steps tracked from pocket	Pocket test

• Sources: Android Central (2025) ⚠️

Fitbit

Fitbit has 144 peer-reviewed studies — more than any other consumer wearable brand.

Peer-reviewed data:

Model	MAPE / Metric	Condition
Sense	~8%	Exercise testing
Charge 2	3.4% diff vs open-source algo	Clinical validation
Charge 2 / Alta	17.1–35.5%	24-hr free-living
Charge / Charge HR	<25% MAPE	20 studies (systematic review)
Two wrist models	Sensitivity >87%, Specificity >97%	Short-bout stepping vs ActiGraph

Fitbit underestimates in lab conditions and overestimates in free-living. A systematic review found it meets acceptable accuracy approximately half the time. Bout detection drops off at >120 steps/min in older adults.

Within-brand variation exists: Classic overestimates, Charge underestimates.

Ankle placement finding: The same Fitbit worn at ankle achieved 5.9% error at 0.4 m/s versus 48–75% error when worn on the wrist — confirming that wrist placement is the limitation, not the algorithm.

No Charge 6 validation has been published.

• Sources: Fuller et al. (2020); Germini et al. (2022); Giurgiu et al. (2023); Straczkiewicz et al. (2023); Delobelle et al. (2024); Nature Scientific Reports (2024)

Samsung Galaxy Watch

Samsung has far fewer validation studies than Apple, Garmin, or Fitbit.

Peer-reviewed data:

Model	Correlation	Condition
Galaxy Watch 4	r=0.82 vs ActivPAL	24-hr free-living

For comparison, Apple Watch 6 achieved r=0.99 vs the same criterion in the same study.

• Sources: Kim et al. (2024)

Consumer testing:

Model	Result	Condition
Galaxy Watch 5 Pro	113 steps off / 6,000 (~1.9% error)	Walk + run
Galaxy Watch 8 Classic	Underperformed expectations	10-watch walk/jog

December 2025 consumer testing noted Samsung "has never done especially well in past step tests." Persistent user reports cite 2,000–3,500+ phantom steps per day during desk work and driving.

• Sources: Android Central (2023, 2025) ⚠️

Oura Ring

Only 2 peer-reviewed studies have examined Oura step accuracy.

Peer-reviewed data:

Model	Metric	Value	Condition
Gen 2/3	MAPE	<10%	Laboratory (combined)
Gen 2/3	Mean difference	+2,124 ± 4,256 steps/day	Free-living (14 days)

Controlled walking accuracy is acceptable; free-living accuracy is not. Hand gestures, cooking, and stirring trigger phantom steps. A 2025 systematic review found 107 smart ring studies (Oura in 72% of them) — almost none examined step accuracy.

• Sources: Kristiansson et al. (2023); Gong & Bang (2025)

Other evidence:

Model	Result
Gen 3 (controlled walk, 5,000 steps)	Within 12 steps
Gen 3 / Ring 4 post "Real Steps" update	~20% decrease in reported steps vs prior algo
Ring 4	Up to 1,000 steps off vs Apple Watch daily

Oura's "Real Steps" update (March 2025) introduced a new ML algorithm for Gen3/Ring 4 to distinguish walking from hand movement. No peer-reviewed validation of the new algorithm exists.

• Sources: Oura Blog (2025); Android Central (2023) ⚠️; Tom's Guide ⚠️

Devices With No or Minimal Peer-Reviewed Step Data

The following devices have zero or near-zero peer-reviewed step validation.

WHOOP

Step counting was added to WHOOP in October 2024. WHOOP claims steps are "validated for on-wrist use" — no published validation exists. WHOOP 5.0/MG reportedly uses a new step algorithm. Steps are positioned as supplementary to Strain.

Polar

Peer-reviewed (older models only):

Model	Finding
A360	"Not valid for any walking condition"

Only older models have been studied. No peer-reviewed step validation exists for current models.

Consumer testing (current models):

Model	Result
Vantage V3	Within 100 steps / 10,000 ("reliable tier")
Vantage M3	+379 extra steps / 10,000 (+3.8%)

Vantage V3 showed significant improvement over older models in December 2025 testing.

• Sources: Henriksen et al. (2022); Roos et al. (2020); Android Central (2025) ⚠️

COROS

Consumer testing only:

Model	Result	Condition
APEX 2 Pro	81 steps off / 5,000 (~1.6% error)	Walk + run
APEX 4	Within 100 steps / 10,000	Walk + jog
APEX 4	~5,041 steps tracked from pocket	Pocket test

APEX 4 improved significantly over APEX 2 Pro. No peer-reviewed step validation exists for any COROS model.

• Sources: Android Central (2023, 2025) ⚠️

Google Pixel Watch

No peer-reviewed step validation exists for Google Pixel Watch.

Wear OS 5.1 (March 2025) introduced an "enhanced step algorithm" — which caused inflated step counts across all models. Google reverted the algorithm in April 2025.

December 2025 testing: Pixel Watch 4 initially performed well, but the Fitbit app stopped mid-test and subtracted approximately 4,000 steps. The device could not be ranked.

• Sources: Google Wear OS 5.1 Release Notes (March–April 2025); Android Central (2025) ⚠️

Why Step Count Accuracy Matters for Correlating Activity and Nutrition

When tracking how food affects your energy and activity levels, step accuracy is foundational. A device that systematically overcounts during desk work will show different patterns than one that undercounts during stroller walks.

Kygo Health integrates with multiple wearable platforms — Oura, Apple Health, Garmin, Fitbit — so you can correlate nutrition with biometrics using whichever device works best for your accuracy profile. Understanding these accuracy characteristics helps you interpret the patterns we surface.

For our full breakdown of wearable accuracy across all biometrics (sleep, HRV, heart rate, SpO2, calories, VO2 max), see: Most Accurate Wearable. Or compare devices yourself with our Wearable Accuracy Comparison Tool.

Important Caveats

1. Lab ≠ Real World. All devices show ~5% MAPE in controlled conditions versus >10% in free-living.

2. Model generation matters. Most studies tested older devices. Current-gen may be better — no studies confirm.

3. Only ~11% of consumer wearables have been validated for any biometric. 249 studies = 3.5% of comprehensive coverage.

4. Algorithms change. Google reverted their step algorithm in April 2025 after overcounting. Oura's "Real Steps" update dropped reported counts ~20%. Your device's accuracy today may differ from what was tested.

5. Individual variation is significant. Gait, arm swing, BMI, age, speed all affect results. Published MAPEs are population averages.

6. Smart rings are understudied for steps. A 2025 systematic review found 107 smart ring studies (72% Oura) — almost all examined sleep or HR, not step accuracy.

7. WellnessPulse and AIM7 are consumer data aggregations, not peer-reviewed research.

8. Study funding matters. Always check who paid.

Key Takeaways

1. Garmin and Apple Watch lead in peer-reviewed accuracy — but all major brands perform acceptably at normal walking speeds.

2. Oura Ring is accurate in controlled conditions but overcounts significantly in real-world use due to hand gestures triggering phantom steps. The March 2025 "Real Steps" update may improve this.

3. WHOOP, COROS, and Google Pixel Watch lack peer-reviewed step validation — use consumer testing data with appropriate skepticism.

4. Free-living accuracy is 2–3x worse than lab accuracy across all devices. Don't expect a study's treadmill results to translate to your workday.

5. What you do matters more than which device you own. Walking speed, arm swing, and device placement affect accuracy more than brand choice.

---

Sources

⚠️ = Consumer testing, not peer-reviewed

1. Fuller D, et al. (2020). JMIR mHealth and uHealth, 8(9), e18694. DOI: 10.2196/18694 — 144 Fitbit, 42 Garmin, 28 Apple studies

2. Germini F, et al. (2022). JMIR, 24(1), e30791. DOI: 10.2196/30791

3. O'Driscoll R, et al. (2024). Sports Medicine. DOI: 10.1007/s40279-024-02077-2 — 24 reviews, 249 studies, 430K participants

4. Kim Y, et al. (2024). Sensors, 24(14), 4658. DOI: 10.3390/s24144658 — 104 adults, 24-hr ActivPAL

5. Kristiansson E, et al. (2023). BMC Medical Research Methodology, 23, 50. DOI: 10.1186/s12874-023-01868-x

6. Choe S & Kang M (2025). Physiological Measurement. DOI: 10.1088/1361-6579/adca82 — 56 studies, 270 effect sizes

7. npj Digital Medicine (2026). DOI: 10.1038/s41746-025-02238-1 — 82 studies, through Series 9/Ultra 2

8. Garmin Validity Review (2020). PMC. DOI: 10.3390/ijerph17134269

9. Feehan LM, et al. (2020). PeerJ. DOI: 10.7717/peerj.9381

10. Roos L, et al. (2020). Int J Environ Res Public Health, 17(20), 7123. DOI: 10.3390/ijerph17207123

11. Henriksen A, et al. (2022). JMIR Formative Research, 6(5), e27248.

12. Giurgiu M, et al. (2023). Technologies, 11(1), 29. DOI: 10.3390/technologies11010029

13. Straczkiewicz M, et al. (2023). JMIR Cancer. DOI: 10.2196/47646

14. Small SR, et al. (2024). Nature Medicine. DOI: 10.1038/s41591-024-03243-w

15. Nature Scientific Reports (2024). DOI: 10.1038/s41598-024-74140-x

16. Niela-Vilén H, et al. (2022). Sensors, 22(7), 2585. DOI: 10.3390/s22072585

17. Modave F, et al. (2017). JMIR mHealth, 5(6), e88. DOI: 10.2196/mhealth.7870

18. Scataglini S, et al. (2025). Int J Obes, 49(4), 541-553. DOI: 10.1038/s41366-024-01659-4

19. Sensors (2025). Sensors, 25(18), 5657.

20. Johnston W, et al. (2021). Br J Sports Med, 55(14), 780-793.

21. Delobelle J, et al. (2024). Digital Health, 10, 20552076241262710. DOI: 10.1177/20552076241262710

22. Gong EJ & Bang CS (2025). Smart ring systematic review. PMC. — 107 studies, Oura in 72%

23. Oura Blog (March 2025). "Real Steps" algorithm update. First-party.

24. Google Wear OS 5.1 Release Notes (March–April 2025). Step algorithm enhancement and revert. First-party.

25. WellnessPulse (2025). ⚠️ Not peer-reviewed

26. ⚠️ Android Central (2023). 6-watch step test.

27. ⚠️ Android Central (December 2025). 10-watch step test.

---

Ready to correlate your activity data with nutrition patterns? Kygo Health integrates with Oura, Apple Health, Garmin, and Fitbit — use whichever device fits your accuracy needs and see how food choices connect to your biometric data.

Download free on iOS or join the Android beta — fully free to join.

---

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 sources or data that should be included here? Reach out at Ryan@kygo.app.