The accuracy of pulse oximeters from fourteen manufacturers was tested during profound brief hypoxic plateaus in 125 subject sets using 50 normal adult volunteers, of whom 29 were studied two to nine times. A data set usually consisted of 10 subjects, and 13 sets were collected between August 1987 and July 1988. In the first 6 sets, six 30-second hypoxic plateaus were obtained per subject at 55 +/- 6% oxyhemoglobin (O2Hb) (range, 40 to 70%). In the last 7 sets, three hypoxic plateaus were obtained at each of four levels, approximately 86, 74, 62, and 50% O2Hb, for the purpose of linear regression analysis. Inspired oxygen was adjusted manually breath by breath in response to arterial oxygen saturation computed on-line from end-tidal oxygen and carbon dioxide tensions. End-plateau arterial blood O2Hb was analyzed by a Radiometer OSM-3 oximeter, and plateau pulse oximeter saturation (SpO2) was read by cursor from a computer record of the analog output. Three to 13 instruments were tested simultaneously by using 1 to 3 duplicate instruments from each of one to seven manufacturers. Variations introduced by manufacturers were tested on subsequent sets in several instruments. An index of error, "ambiguity" (alpha) of oxygen saturation, was defined as the absolute sum of bias and precision (mean and SD of SpO2 - O2Hb) at O2Hb = 55.8 +/- 4.5%, preserving the sign when bias was significant at P less than 0.05. Ambiguity values for finger probes (unless specified) with latest data were: Physio-Control, 3.9 (ear, 3.3); Puritan-Bennett, -4.4; Criticare, 5.8 (forehead, 4.7); Kontron, 5.9 (infant probe) and 6.1 (ear, 5.8; forehead, 7.1); Biochem, -6.0; Datex 6.4 (ear, 6.9; forehead, 6.8); Critikon, 8.4; SiMed, 8.6; Marquest, 9.0; Novametrix, 10.2; Invivo, -12.2 (ear, -14.3); Nellcor, -15.1; Ohmeda, -21.2; and Radiometer, -21.2 (ear, -9.6). Linear regression slopes of 36 instruments from twelve manufacturers generally deviated from 1 in proportion to alpha. The data showed substantial differences in bias and precision between pulse oximeters at low saturations, the most common problems being underestimation of saturation and failing precision.