This study was designed to discriminate, according to their accuracy, between eight pulse oximeters (PO) using a realistic evaluation performed by means of a moderate exercise to provide artefacts and thus represent a typical clinical situation. Accuracy was assessed by calculating error on saturation (EOS) as the difference between simultaneous measurement of non-invasive (SpO2) and invasive functional (SfO2) arterial blood oxygen saturation. An index of homogeneity of EOS (InH), calculated using a Lehmann analysis, was chosen as a representative of error dispersion, leading to a statistical classification; the influence on error introduced by the drift of EOS with SfO2 was thus eliminated. We obtained a clear-cut separation of the eight PO (Datex Satlite (DA), Kontron 7840 (KO), Nellcor 200 (NE), Novametrix 505 (NO), Ohmeda 3700 (OH), PPG-Hellige (PP), Physiocontrol 1600 (PH), and Radiometer Oxi (RA] into two statistically different subgroups by comparing their InH standard deviation (SD). There is a highly statistically significant difference between the accuracy measured for the high precision subgroup with an InH SD ranging from +/- 1.6 to +/- 1.9% SaO2 (DA, PH, and RA), and that measured for the low precision subgroup whose InH SD ranges from +/- 2.8 to +/- 5.0% SaO2 (NE, OH, NO, KO, PP). These results suggest that, in our clinical conditions, there is a wide range of inhomogeneity of error from one PO to another. The error introduced by the drift of EOS with SfO2 should be considered separately.