Arterial hypoxemia has been reported in horses during heavy exercise, but its mechanism has not been determined. With the use of the multiple inert gas elimination technique, we studied five horses, each on two separate occasions, to determine the physiological basis of the hypoxemia that developed during horizontal treadmill exercise at speeds of 4, 10, 12, and 13-14 m/s. Mean, blood temperature-corrected, arterial PO2 fell from 89.4 Torr at rest to 80.7 and 72.1 Torr at 12 and 13-14 m/s, respectively, whereas corresponding PaCO2 values were 40.3, 40.3, and 39.2 Torr. Alveolar-arterial PO2 differences (AaDO2) thus increased from 11.4 Torr at rest to 24.9 and 30.7 Torr at 12 and 13-14 m/s. In 8 of the 10 studies there was no change in ventilation-perfusion (VA/Q) relationships with exercise (despite bronchoscopic evidence of airway bleeding in 3) and total shunt was always less than 1% of the cardiac output. Below 10 m/s, the AaDO2 was due only to VA/Q mismatch, but at higher speeds, diffusion limitation of O2 uptake was increasingly evident, accounting for 76% of the AaDO2 at 13-14 m/s. Most of the exercise-induced hypoxemia is thus the result of diffusion limitation with a smaller contribution from VA/Q inequality and essentially none from shunting.