We employed an associative learning paradigm to test the hypothesis that exercise hyperpnea in humans arises from learned responses forged by prior experience. Twelve subjects undertook a "conditioning" and a "nonconditioning" session on separate days, with order of performance counterbalanced among subjects. In both sessions, subjects performed repeated bouts of 6 min of treadmill exercise, each separated by 5 min of rest. The only difference between sessions was that all the second-to-penultimate runs of the conditioning session were performed with added dead space in the breathing circuit. Cardiorespiratory responses during the first and last runs (the "control" and "test" runs) were compared for each session. Steady-state exercise end-tidal PCO₂ was significantly lower (P = 0.003) during test than during control runs for both sessions (dropping by 1.8 ± 2 and 1.4 ± 3 Torr during conditioning and nonconditioning sessions, respectively). This and all other test-control run differences tended to be greater during the first session performed regardless of session type. Our data provide no support for the hypothesis implicating associative learning processes in the ventilatory response to exercise in humans.