There is a growing need to measure arterial oxygen saturation with a non-invasive method during heavy exercise under severe hypoxic conditions. Although the accuracy of pulse oximetry has been challenged by several authors, it has not been done under extreme conditions. The purpose of this study was to evaluate the accuracy of a pulse oximeter (Satllite. Datex, Finland) during exercise under hypoxic conditions where arterial oxygen saturation was below 75%, simulating exercise at extreme altitude. Ten healthy non-smoking men performed two exercise studies of 30 min under normoxia and under hypoxia on two consecutive days. The exercise intensity was 80% of maximal O2 consumption of VO2max. Arterial oxygen saturation measured by pulse oximetry was corrected (SpO2[corr]) according to previously published equations and was compared to arterial oxygen saturation (SaO2) in blood samples taken simultaneously from the radial artery. Reference arterial saturation values ranged from 57.2 to 97.6% for the whole data set. This data set was split according to low (SaO2 < or = 75%) and high (SaO2 > 75%) SaO2 values. The error of pulse oximetry (SpO2[corr]-SaO2) was 2.05 (0.87)% [mean (SD)] and 1.80 (1.81)% for high and low SaO2 values, respectively. SpO2[corr] and SaO2 were highly correlated (r = 0.93, SEE = 1.81) for low values. During high-intensity constant workload under severe hypoxic conditions, once corrected, pulse oximetry provides an estimate of SaO2 with a mean error of 2%. Thus, the correction previously described for SpO2 values above 75% saturation applies also to SpO2 values in the range of 57-75% during exercise under hypoxic conditions.