A Blood Oxygen Reading of 90 While Sleeping: The Number and the Sensor's Own Margin of Error
A reading this specific deserves a specific answer, and the most honest one starts with how much error the device itself is allowed to carry before a single number means anything.
Evidence rating for the proposition “Pulse oximeter measurement bias varies systematically with skin pigmentation”: Uncertain · KnowYourPrime Evidence Registry, Finding KYP-ER-0002
A blood oxygen reading of 90 while sleeping is worth paying attention to, but it is an estimate from a wrist sensor, not a measurement of the oxygen in your arterial blood. Pooled across the studies that tested Apple Watch against a reference, the average error was close to zero, yet individual readings ran from about 4 points below to about 4 points above the reference. A single 90 carries that spread; a 90 that repeats, or comes with symptoms, carries more weight.
The Accuracy Bar for Pulse Oximeters, and What Has Been Measured on a Watch
Medical pulse oximeters are evaluated against an accuracy standard set by the American National Standards Institute, the Association for the Advancement of Medical Instrumentation and the International Electrotechnical Commission. It requires a root-mean-square accuracy error of 3 percent or less against direct arterial blood measurement.
Even conventional pulse oximeters miss that bar in pooled data. A systematic review of 23 pulse oximetry studies, nearly 60,000 participants and more than 197,000 readings paired with arterial blood, found root-mean-square errors of 3.96, 4.71 and 4.15 percent across light, medium and dark skin pigmentation. Those studies tested clinical pulse oximeters, not watches: the review's only wearable data was pulse rate. Whether the reading works the same for everyone is the narrower question this research speaks to directly.
The closest evidence for a watch is a living systematic review of Apple Watch validation studies, 82 studies across 14 health metrics, of which the blood oxygen studies are a subset. Its pooled blood oxygen readings had almost no average bias (-0.04 percentage points), but the 95 percent limits of agreement ran from -4.00 to +3.94 percentage points. Those figures describe Apple Watch; other brands use different sensors and algorithms, and the same numbers do not carry over to them.
2 studies
- In a living review of 82 Apple Watch validation studies across 14 health metrics, the pooled blood oxygen studies had a mean bias of -0.04 percent (95% CI -0.42 to 0.35) with 95 percent limits of agreement from -4.00 to 3.94 percentage points.
- Pooling 23 studies of conventional pulse oximeters (59,684 participants, 197,353 readings paired with arterial blood), root-mean-square accuracy error was 3.96, 4.71 and 4.15 percent for light, medium and dark skin pigmentation, all above the 3 percent threshold. The review's wearable data covered pulse rate only.
Correction, 2026-10-14: An earlier version of this article said a meta-analysis of 23 studies found wearable devices miss the 3 percent accuracy standard, and used its 3.96 to 4.71 percent errors as the margin on a watch reading. Those 23 studies tested conventional pulse oximeters against arterial blood, not watches. The watch figures now come from a pooled analysis of Apple Watch studies. Updated 2026-10-15: that analysis pooled the blood oxygen studies within an 82-study review covering 14 health metrics, not 82 blood oxygen studies, and the wording now says so.
What a Few Points of Error Actually Does to the Number 90
Limits of agreement describe where 95 percent of individual readings fell relative to the reference across the pooled studies. On the Apple Watch figures, a displayed 90 is consistent with a reference value somewhere around 86 to 94.
That spread is a property of many readings across many people, not a guaranteed margin on one wrist on one night, and it runs in both directions. It does not make 90 meaningless. It makes a single reading a starting point for attention rather than a measurement to treat as if it came from a hospital monitor.
These figures pool Apple Watch studies across models, settings and people. They describe the spread of individual readings in those studies, not the error on any particular device or night, and they do not transfer to other brands.
When a Reading of 90 Is a Factual Reason to Seek Medical Care
Device error works in both directions, it cuts against complacency as much as it cuts against panic. A reading at or below 90 that repeats across more than one night, or that shows up alongside shortness of breath, chest pain, confusion, or bluish lips or fingertips, is a factual reason to seek medical evaluation. Not a reason to wait and recheck the watch.
A single reading with no symptoms and no repeat occurrence is the situation this device-error evidence speaks to most directly. Something sustained or symptomatic is a different situation, and it should be treated as one.
Common questions
Is a blood oxygen reading of 90 on a smartwatch accurate?
It is an estimate. Pooled across the blood oxygen studies in a living review of Apple Watch validation research, readings had almost no average bias, but individual readings ran from about 4 points below to about 4 points above the reference, so a displayed 90 is consistent with a true value of roughly 86 to 94.
Could the true reading be higher or lower than what the watch shows?
Either. The documented error runs in both directions, so a displayed 90 could reflect a true value that is somewhat higher or somewhat lower, not a fixed, exact number.
Does device error mean a low reading can be ignored?
No. It means a single reading with no symptoms is weaker evidence than it looks. A reading that repeats across nights or comes with symptoms like shortness of breath or chest pain is a different situation and a factual reason to seek medical care.
Why do smartwatches have this error in the first place?
Wearables estimate blood oxygen optically through the skin rather than sampling arterial blood directly. Even conventional pulse oximeters, which use the same light-based method, miss the 3 percent accuracy standard in pooled data, so some error is built into the approach before a watch is involved.
KnowYourPrime Evidence Registry: Finding KYP-ER-0017: Can Blood Oxygen Monitoring Rule Out Sleep Apnea? · Finding KYP-ER-0001: Smartwatch Blood Oxygen (SpO2) Accuracy Against Arterial Blood (SaO2) · Finding KYP-ER-0019: Blood Oxygen Readings With Poor Circulation · Finding KYP-ER-0023: Apple Watch Blood Oxygen Accuracy Against Medical Devices
Sources
- The accuracy of Apple Watch measurements: a living systematic review and meta-analysis.
- Impact of Skin Pigmentation on Pulse Oximetry Blood Oxygenation and Wearable Pulse Rate Accuracy: Systematic Review and Meta-Analysis [author affiliations include Verily Life Sciences]
- Low Perfusion and Missed Diagnosis of Hypoxemia by Pulse Oximetry in Darkly Pigmented Skin: A Prospective Study
- Effects of skin pigmentation on pulse oximeter accuracy at low saturation
- Pulse oximetry: principles and limitations