To further our understanding of the best way to measure regional CBF with positron emission tomography (PET), we directly compared two candidate tracers ([15O]water and [11C]butanol, administered intravenously) and two popular implementations of the one-compartment (1C) model: the autoradiographic implementation representing a single PET measurement of tissue radioactivity over 1 min and a dynamic implementation representing a sequence of measurements of tissue radioactivity over 200 s. We also examined the feasibility of implementing a more realistic, and thus more complex, distributed-parameter (DP) model by assigning fixed values for all of its parameters other than CBF and tracer volume of distribution (Vd), a requirement imposed by the low temporal resolution and statistical quality of PET data. The studies were performed in three normal adult human subjects during paired rest and visual stimulation. In each subject seven regions of interest (ROIs) were selected, one of which was the primary visual cortex. The corresponding ROI were anatomically equivalent in the three subjects. Regional CBF, Vd, tracer arrival delay, and dispersion were estimated for the dynamic data curves. A total of 252 parameter sets were estimated. With [11C]butanol both implementations of the 1C model provided similar results (r = 0.97). Flows estimated using the 1C models were lower (p < 0.01) with [15O]water than with [11C]butanol. In comparison with the 1C model, the constrained version of the DP used in these studies performed inadequately, overestimating high flow and underestimating low flow with both tracers, possibly as the result of the necessity of assigning fixed values for all of its parameters other than CBF and Vd.