This study examined the dynamics of the cerebral blood flow response to hypoxia and hypercapnia in humans. Middle cerebral artery blood flow (MCAF) was assessed continuously using transcranial Doppler ultrasound. MCAF was calculated on a beat-by-beat basis as the product of the intensity-weighted mean velocity and the total power of the reflected signal. End-tidal PCO2 (PETCO2) and PO2 (PETO2) were controlled using a dynamic end-tidal forcing system. Six repeats of each of four protocols were administered to six subjects. The first was a control protocol with PETO2 held at 100 Torr and PETCO2 held 1-2 Torr above eucapnia throughout. The second was a hypoxic step protocol with PETO2 lowered from control values to 50 Torr for 20 min. The third was a hypercapnic step protocol with PETCO2 elevated from control by 7.5 Torr for 20 min. The fourth was a hypoxic-and-hypercapnic step protocol lasting 20 min. The total power of the Doppler signal remained relatively constant, suggesting that the cross-sectional area of the vessel changed little. After the initial transient in MCAF at the onset of the stimulus, no adaptation or progressive increase was observed over the remaining 20 min. A simple model consisting of a single pure delay, gain terms, time constants, and offsets for the on and off transients was fitted to the hypoxic and hypercapnic protocols. For hypercapnia, all the parameters for the onset were significantly different from the relief of the stimulus. The asymmetry was characterized by a slower on transient than off transient and also by a degree of undershoot after the relief of hypercapnia. Finally, the results from this study show that the cerebral blood flow response to hypoxia and hypercapnia in humans is much faster than has previously been thought.