A pulse oximeter reading is an estimate, not a measurement — and the size of the error is not the same for everybody. Here is what the number means, how wide the margin really is, and the accuracy gap across skin tones that regulators are still working to close.
A pulse oximeter shines red and infrared light through tissue and infers saturation from how much of each colour is absorbed. Oxygenated and deoxygenated haemoglobin absorb the two wavelengths differently, so the ratio gives an estimate. The clinical gold standard, an arterial blood gas, measures saturation directly from a blood sample — written SaO2. SpO2 is the estimate of SaO2. Every limitation below comes from that gap.
| Reading | What it means | What to do |
|---|---|---|
| 96% or more | Normal range | Nothing |
| 95% | Acceptable — keep an eye on it | Continue monitoring |
| 93–94% | Below the normal range | Contact your GP |
| 92% or less | Low — needs assessment now | Emergency care |
The whole clinically interesting range — normal to emergency — spans about four percentage points. The FDA's own margin on a single cleared-device reading is roughly the same width: a reading of 90% generally means a true saturation somewhere between 86% and 94%.1 That is why no clinician acts on one number in isolation, and neither should you.
In 2020, researchers compared paired pulse-oximeter and arterial blood-gas readings and looked for occult hypoxemia — true arterial saturation below 88% while the oximeter was still showing a reassuring 92–96%. The pattern held across two separate cohorts:3
A later systematic review pooled 263,000 paired observations and found the same direction of error. Oximeters overestimated saturation by an average of 1.54 percentage points in Black patients versus 0.62 in White patients, and pooled occult-hypoxemia prevalence was 11.4% versus 6.5% — a prevalence ratio of 1.71, or 71% higher. The reviewers rated the evidence for overestimation in Black patients as moderate strength, and evidence on downstream clinical outcomes as insufficient — the accuracy gap is well established; how much harm it causes is still being quantified.4
The bias grows as saturation falls. Laboratory work found a maximum bias of about 3.6 percentage points in darkly pigmented subjects in the 60–70% saturation range, but under 1 point above 80% saturation.7 In other words, the error is largest exactly where the stakes are highest — critical care — and smallest in the healthy range where a smartwatch spends its time. That does not make it irrelevant to consumers: it means a wrist reading of 94% deserves the same scepticism from everyone, and a borderline reading deserves more follow-up, not less, if you have darker skin.
Under the long-standing approach, a device could be validated on as few as ten people, with darkly pigmented subjects making up 15% of test subjects, or two people, whichever was greater.7 In January 2025 the FDA issued draft guidance — Pulse Oximeters for Medical Purposes — proposing a substantially larger and more representative evidence base: around 150 participants, with skin tone recorded using both a subjective scale (the Monk Skin Tone Scale, 10 values, analysed in three groups: 1–4, 5–7 and 8–10) and an objective optical measure (individual typology angle).5,8
As of this guide's last update the FDA document remains draft guidance — non-binding recommendations, not a rule in force. Devices cleared before it takes effect were validated under the older standard. If a manufacturer claims comparable performance across skin tones, ask what the labelling actually says.
Two things are worth saying plainly about this. First, the spread between the best and worst device was more than twofold — "has an SpO2 sensor" tells you almost nothing about accuracy. Second, this particular study found skin tone had little correlation with error in its sample; that is one study at healthy saturations and does not overturn the clinical literature above, but it is what the data showed and it belongs in the record.
1. Sit still and rest your arm for a minute first — motion is the biggest single source of junk readings. 2. Warm cold hands. 3. Wear the band snug and one finger-width above the wrist bone. 4. Take two or three readings and look at the pattern, not the first number. 5. Write down what you saw and when, so a clinician has a trend rather than an anecdote.