Blood oxygen may be the least understood number your watch collects. The textbook answer is easy: 95 to 100 percent at rest, near sea level.
The harder and more useful question is how much to trust any particular reading, because the honest answer is less than the two decimal places imply, and not equally for everyone.
Key takeaways
Normal resting oxygen saturation is 95 to 100 percent; below 90 percent is the clinical definition of hypoxemia (StatPearls).
Consumer and medical devices both overestimate. In controlled testing, a smartwatch fell within 2 percent of true saturation only 32 percent of the time, against 49 percent for a fingertip oximeter (Jiang et al. 2026, JMIR Formative Research).
Pulse oximeters miss low oxygen more often in patients with darker skin. Across 732,505 paired measurements, undetected low oxygen was 1.67 times more common in Black than White hospitalised patients (Parr et al. 2024, Journal of General Internal Medicine).
Accuracy is worst exactly where it matters most, at low saturations.
What does SpO2 actually measure?
The percentage of your haemoglobin currently carrying oxygen.
Red blood cells ferry oxygen from your lungs to the rest of you. A reading of 97 percent means 97 of every 100 available seats are occupied. Because your heart and lungs top the supply up continuously, saturation stays remarkably stable in healthy people, whether you are reading a book or climbing stairs. That stability is what makes a genuine change meaningful, and also what makes a wobbly reading obviously suspect.
Normal resting values sit at 95 to 100 percent near sea level. Hypoxemia is defined as saturation below 90 percent, or an arterial oxygen pressure below 60 mmHg (StatPearls).
Note the size of that scale. The entire useful range is about ten percentage points wide, and the difference between fine and worth-a-conversation is around five. That is a narrow target for a sensor reading light bounced through skin.
How accurate is the reading on my watch?
Less accurate than it looks, and biased in a specific direction.
A validation study induced controlled low oxygen and compared devices against arterial blood, the reference standard. Both an Apple Watch Series 7 and a Masimo fingertip oximeter consistently overestimated saturation (Jiang et al. 2026):
Within 2% of true value | Error at saturations below 88% | |
|---|---|---|
Fingertip oximeter | 49.0% of readings | 3.52% |
Smartwatch | 32.1% of readings | 5.82% |
Two things follow. The watch was within two points of the truth about a third of the time. And accuracy got worse as saturation fell, which is precisely backwards from what you would want. The authors noted the smartwatch's bias makes it likely to miss genuinely low oxygen sitting just below 88 percent. Both devices exceeded the error thresholds the FDA sets for clearance.
This is not a reason to ignore the number. It is a reason to read it as a trend against your own baseline rather than as a measurement, which is the same logic that applies to walking heart rate, where the sensor is similarly imperfect but consistently so.
Does skin tone affect the reading?
Yes, and this is the most consequential thing to know about pulse oximetry.
Pulse oximeters work by shining light through tissue, and skin pigmentation changes how that light behaves. A systematic review pooled 15 studies covering 732,505 paired oximetry measurements from 207,464 hospitalised patients and looked at occult hypoxemia: cases where the oximeter read normal while arterial blood showed the patient was genuinely low (Parr et al. 2024).
Compared with White patients, occult hypoxemia was:
1.67 times more common in Black patients
1.39 times more common in patients identifying as Asian, Latinx, Indigenous, multiracial or another race or ethnicity
The reviewers also found that Black patients with undetected low oxygen appeared to receive poorer treatment as a result.
Two boundaries keep this accurate. All of that evidence comes from hospitalised patients, and the review found no evidence at all from outpatient settings, so the size of the effect in everyday consumer use is genuinely unknown. And the studies relied on self-identified race, which is a rough proxy for skin pigmentation.
The practical implication is not that the number is useless. It is that a normal-looking reading is weaker reassurance for some people than for others, and that symptoms should always outrank a device.
What should I do with a low reading?
Retake it properly before you interpret it, because most surprising readings are measurement problems.
Sit still for two or three minutes first. A reading taken straight after moving reflects the movement.
Warm your hands. Cold fingers and wrists carry less blood flow, which weakens the signal.
Keep the band snug, about a finger-width above the wrist bone. A loose band is the single most common cause of a bad wrist reading.
Breathe normally. No deep dramatic breaths; you want your ordinary resting picture.
Take three readings a minute apart and trust the cluster, not the outlier. If they refuse to settle, try again later rather than forcing it.
Stay with one device. Comparing a watch against a fingertip clip compares methods, not health.
Fingertip clips generally give the steadiest signal, which is why clinics use them, though nail polish and acrylics interfere. Wrist wearables trade precision for something genuinely valuable: effortless readings around the clock, which is what builds a trend worth reading.
When is it worth a conversation?
The number alone is a weak signal. The number plus symptoms is a strong one.
Given how often devices overestimate and how much worse they perform at low values, breathlessness, confusion, chest pain or blue-tinged lips matter more than a reassuring percentage. Do not let a normal reading talk you out of symptoms, particularly given the skin tone findings above.
What is worth raising with a clinician is a sustained change: readings consistently below your usual range across many days rather than one odd afternoon. Overnight patterns have their own explanation, which we cover in why SpO2 dips at night and at altitude.
Frequently asked questions
What is a normal blood oxygen level?
For healthy adults at or near sea level, 95 to 100 percent at rest. Saturation below 90 percent is the clinical definition of hypoxemia (StatPearls). Readings run slightly lower during sleep and at altitude, and both are expected.
Is 94 percent oxygen level okay?
It sits just below the usual reference range, but a single reading is weak evidence either way given that devices fall within 2 percent of the truth only a third to half the time (Jiang et al. 2026). Retake it sitting still with warm hands and a snug band. A sustained shift below your own usual range, or any reading alongside breathlessness, is worth a clinician's view.
Why is my Apple Watch oxygen reading different from my finger oximeter?
Because they measure differently and both have error. In controlled testing both overestimated, with the fingertip device within 2 percent of the true value 49 percent of the time against 32 percent for the smartwatch (Jiang et al. 2026). Pick one device and follow its trend rather than reconciling the two.
Are pulse oximeters less accurate on darker skin?
The hospital evidence says yes. Across 732,505 paired measurements, undetected low oxygen was 1.67 times more common in Black patients and 1.39 times more common in several other groups compared with White patients (Parr et al. 2024). No equivalent evidence exists from everyday outpatient use, so the everyday size of the effect is unknown. Symptoms should outrank a normal reading.
How can I get a more reliable reading at home?
Rest for two to three minutes, warm your hands, keep the band snug and a finger-width above the wrist bone, breathe normally, and take three readings a minute apart. Trust the cluster rather than any single value, and measure at a similar time of day so the long-term trend stays comparable.
Related reading
Overnight and altitude readings have their own explanations, covered in why SpO2 dips at night and at altitude. For the other overnight signal, which is steadier and measured quite differently, see your breathing rate while you sleep.
Knit shows patterns, not diagnoses. Talk to a clinician about readings that concern you.
