OBJECTIVE To evaluate the performance of two pulse oximeters in the measurement of arterial hemoglobin saturation in hypoxemic children. METHODS Prospective, repeated-measures observational study. METHODS A 16-bed pediatric intensive care unit in a children's tertiary hospital. METHODS Sixty-six patients with arterial saturation of <90%. METHODS Three arterial blood samples were taken from each subject during a 48-hr period. Pulse oximeter measurements of arterial saturation were compared with arterial saturation determined by cooximetry. RESULTS Arterial saturation was measured using one or both pulse oximeters (SpO2) and compared with the arterial hemoglobin saturation determined by cooximetry (SaO2). Sixty-two subjects were studied, using the Ohmeda pulse oximeter giving 185 data points (78 with saturations <75% [defined by the average of pulse oximeter and cooximeter]); 53 subjects were studied, using the Hewlett-Packard pulse oximeter yielding 155 data points (60 with saturations <75%). SpO2 ranged from 24% to 94%. Bias and precision of the Ohmeda pulse oximeter were -2.8% and 4.8% >75% and -0.8% and 8.0% <75%. Bias and precision of the Hewlett-Packard pulse oximeter were -0.5% and 5.1% >75% and 0.4% and 4.6% <75%. Intrapatient regression coefficient (r) for the differences between pulse oximeter and cooximeter was 0.58 for the Ohmeda and 0.59 for the Hewlett-Packard. Regression coefficients for predicting change in cooximeter value given a change in the Ohmeda pulse oximeter were 0.59 and 0.71 <75% and >75%, respectively. Similar coefficients for the Hewlett-Packard pulse oximeter were 0.50 and 0.70, respectively. CONCLUSIONS The performance of the Ohmeda pulse oximeter deteriorated below an SpO2 of 75%. The Hewlett-Packard pulse oximeter performed consistently above and below an SpO2 of 75%. The ability of both pulse oximeters to reliably predict change in SaO2 based on change in pulse oximetry was limited. We recommend measurement of PaO2 or SaO2 for important clinical decisions.