A retrospective evaluation of simultaneous tests of oximeters of various manufacturers in volunteer subjects disclosed greater errors at low saturations in subjects with low hemoglobin (Hb) concentrations. Forty-three pulse oximeters of 12 manufacturers studied over a period of 10 months showed that, at a mean arterial oxygen saturation (SaO2) level of 54.5%, as Hb concentration fell, average pulse oximeter (SpO2) bias increased approximately linearly from 0 at Hb greater than 14 g/dl to about -14% at 8 less than Hb less than 9 g/dl. At SaO2 = 53.6%, the mean bias (SaO2--SpO2) of 13 oximeters of 5 manufacturers averaged -15.0% (n = 43) in a subject with Hb = 8 g/dl, but -6.4% (n = 390) in nonanemic subjects. The additional bias in the anemic subject increased with desaturation. It was 0.13% at SaO2 = 98.5% (n = 13), -1.31% at 87.5% (n = 38), -2.71% at 75.1% (n = 38), -5.18% at 61.3% (n = 26), and -9.95% at 53.6% (n = 41); n is the product of the number of oximeters and number of tests in each saturation range. The instruments that showed the greatest errors at low saturations in nonanemic subjects also showed the greatest additional errors associated with anemia (the range between manufacturers of anemic incremental error at about 53% being from -3.2 to -14.5%) and conformed well to the relationship bias (anemic) = 1.35 x bias (normal) -8.18% (r = 0.94; Sy.x = 3.3%). The error due to anemia was zero at 97% SaO2 and became evident when SaO2 fell below 75%.