1. The pattern of breathing during the approach to the steady state following step changes of end-tidal P(CO2) and P(O2) has been determined in normal conscious human subjects. Three types of step were studied: (a) steps of P(A, CO2) against a constant background of hyperoxia (P(A, O2) approximately 200), an almost pure intracranial chemoreceptor stimulus, (b) steps of P(A, O2) between approximately 50 and 80 torr against a background of constant mild hypercapnia, an arterial chemoreceptor stimulus, and (c) steps of P(A, CO2) against a background of constant hypoxia (P(A, O2) approximately 50), a mixed stimulus. Steps were small and the responses barely detectable by the subjects.2. Steps of CO(2) in hyperoxia produced the slowest approach to the steady state. A single exponential fitted the ventilation response up to about 4 min (mean half time 83 sec for the ;up' and 69 sec for the ;down' transients). During the transient the pattern of change of tidal volume (V(T)) and expiratory time (T(E)) was the same as in the steady state. Inspiratory time (T(I)), however, in the early part of the transient, changed in the opposite direction to T(E), returning to its steady value only after 1(1/2)-3 min. This effect occurred in both ;up' and ;down' transients and resulted in a smaller change of respiratory frequency than would have been predicted from the steady-state response.3. Hypoxic steps produced the fastest approach to the steady state with mean half-times for ventilation of 10.9 sec for the ;up' transients and 6.6 sec for the ;down'. T(I) followed the same pattern during the transient as in the steady state, whereas T(E), following the step out of hypoxia, lengthened to far beyond its final steady value within five breaths of the step, only returning to its steady-state value 3-4 min after the step. This resulted in an exaggerated change of frequency during the early part of the transient.4. Steps of CO(2) in hypoxia, a mixed peripheral and central chemoreceptor stimulus, showed a ventilation response which was best fitted by two exponentials, the half-times of which were consistent with those obtained for the separate responses. The patterning was also consistent with a mixed response, more so for T(I) than for T(E).5. The steady-state pattern derived from the pre-switch means was consistent with the pattern previously described.6. Possible mechanisms are discussed. It is suggested that these results could explain the different patterns seen in the past by those using re-breathing and steady-state techniques.7. The validity of using one or two breath oxygen or nitrogen tests (or other similar tests) as a quantitative measure of the hypoxic response in man is questioned.