1. Recently it was suggested that submaximal cardiac output (Q) could vary in response to changes in arterial O2 concentration (Ca,O2), so that arterial O2 delivery (Qa,O2 = Q x Ca,O2, in ml min-1) is kept constant. 2. This hypothesis was tested on eight healthy male subjects, at rest and during exercise (50, 100 and 150 W) in three conditions: normaemia (N), after 6 weeks of endurance training (T), and 2 days after subsequent autologous blood reinfusion (P). 3. Measured variables were oxygen consumption (VO2), by open circuit method, Q, by a CO2 rebreathing method, and haemoglobin concentration ([Hb]), by a photometric method. Ca,O2 was calculated as the product of [Hb], arterial O2 saturation (0.97), and the O2 binding coefficient. 4. [Hb] and thus Ca,O2 increased by 2.6% (T vs. N) and subsequently by further 5.8% (P vs. T). VO2 and Qa,O2 were linear functions of power (w), both relationships being unaffected by changes in Ca,O2. As a consequence, the linear Q vs. VO2 relationships were shifted downward as Ca,O2 increased. 5. The VO2 vs. w and the Qa,O2 vs. w relationships had the same slope. Therefore, the difference between Qa,O2 (w) and VO2 (w), equal to O2 flow in mixed venous blood (Qv,O2), was constant. 6. In conclusion, the tested hypothesis was supported by the present results. The observed constancy of Qv,O2 suggested that Qv,O2 may play a key role in regulating the cardiovascular response to exercise.