We studied the performance of a portable pulse oximeter in 123 consecutive adult patients by spot-checking with a finger probe and by spectrophotometry of oxygen saturation on a simultaneous arterial blood sample. 88 patients were overtly hypoxemic (HbO2 less than 90%) and 26 showed severe hypoxemia (HbO2 36-70%). The differences between the two methods showed a skewed distribution with a positive tail due to the over-estimation of lower saturation values by the pulse oximeter. Overall, the 95% confidence interval for the median difference ranged from -0.6 to +0.5%. The limits of agreement (distribution-free 95% confidence interval for the sample) were -5.8 to +11.6%. Pulse oximetry can be recommended as a first assessment of the respiratory balance only if a cut-off value of HbO2 equal to 90% in nonsmoking, air-breathing subjects is acceptable. The finger probe implies a response delay of approximately 30 s, making the instrument rather insensitive to short hypoxemic transients. With a predictive value around 90%, the pulse oximeter may be a useful portable screening tool.