Oxygen saturation, SpO2%, was recorded during rapidly induced 42.5 +/- 7.2-s plateaus of profound hypoxia at 40-70% saturation by 1 or 2 pulse oximeters from each of six manufacturers (NE = Nellcor N100, OH = Ohmeda 3700, NO = Novametrix 500 versions 2.2 and 3.3 (revised instrumentation), CR = Criticare CSI 501 + version .27 and version .28 in 501 & 502 (revised instrumentation), PC = PhysioControl Lifestat 1600, and MQ = Marquest/Minolta PulseOx 7). Usually, one probe of each pair was mounted on the ear, the other on a finger. Semi-recumbent, healthy, normotensive, non-smoking caucasian or asian volunteers (age range 18-64 yr) performed the test six to seven times each. After insertion of a radial artery catheter, subjects hyperventilated 3% CO2, 0-5% O2, balance N2. Saturation ScO2, computed on-line from mass spectrometer end-tidal PO2 and PCO2, was used to manually adjust FIO2 breath by breath to obtain a rapid fall to a hypoxic plateau lasting 30-45s, followed by rapid resaturation. Arterial HbO2% (Radiometer OSM-3) sampled near the end of the plateau averaged 55.5 +/- 7.5%. ScO2% (from the mass spectrometer) and SaO2% (from pH and PO2, by Corning 178) differed from HbO2% by + 0.2 +/- 3.6% and 0.4 +/- 2.8%, respectively. The mean and SD errors of pulse oximeters (vs. HbO2%) were: (table; see text) The plateaus were always long enough to permit instruments to demonstrate a plateau with ear probes, but finger probes sometimes failed to provide plateaus in subjects with peripheral vasoconstriction. Nonetheless, SpO2 read significantly too low with finger probes at 55% mean SaO2.(ABSTRACT TRUNCATED AT 250 WORDS)