Pulse oximetry has been proposed as a noninvasive continuous method for transcutaneous monitoring of arterial oxygen saturation of hemoglobin (tcSO2) in the newborn infant. The reliability of this technique in detecting hyperoxemia is controversial, because small changes in saturation greater than 90% are associated with relatively large changes in arterial oxygen tension (PaO2). The purpose of this study was to assess the reliability of pulse oximetry using an alarm limit of 95% tcSO2 in detecting hyperoxemia (defined as PaO2 greater than 90 mm Hg) and to examine the effect of varying the alarm limit on reliability. Two types of pulse oximeter were studied alternately in 50 newborn infants who were mechanically ventilated with indwelling arterial lines. Three arterial blood samples were drawn from every infant during routine increase of inspired oxygen before intratracheal suction, and PaO2 was compared with tcSO2. The Nellcor N-100 pulse oximeter identified all 26 hyperoxemic instances correctly (sensitivity 100%) and alarmed falsely in 25 of 49 nonhyperoxemic instances (specificity 49%). The Ohmeda Biox 3700 pulse oximeter detected 13 of 35 hyperoxemic instances (sensitivity 37%) and alarmed falsely in 7 of 40 nonhyperoxemic instances (specificity 83%). The optimal alarm limit, defined as a sensitivity of 95% or more associated with maximal specificity, was determined for Nellcor N-100 at 96% tcSO2 (specificity 38%) and for Ohmeda Biox 3700 at 89% tcSO2 (specificity 52%). It was concluded that pulse oximeters can be highly sensitive in detecting hyperoxemia provided that type-specific alarm limits are set and a low specificity is accepted.