The effects of tissue heterogeneity on the estimation of regional cerebral glucose utilization (rCMRglc) in normal humans with [18F]2-fluoro-2-deoxy-D-glucose ([18F]FDG) and positron emission tomography (PET) were compared with respect to the various kinetic models of the [18F]FDG method. The kinetic models were conventional homogeneous tissue models of the [18F]FDG method, with (4K Model) and without (3K Model) a rate constant to account for an apparent loss of [18F]2-fluoro-2-deoxy-D-glucose-6-phosphate ([18F]FDG-6-P), and a tissue heterogeneity model (TH Model). When either of the kinetic models designed for homogeneous tissues was applied to heterogeneous tissues, estimates of the rate constant for efflux of [18F]FDG from the tissue (k2*) and of the rate constant for phosphorylation of [18F]FDG (k3*) decreased as the duration of the experimental period was increased. When the 4K Model was used, estimates of the rate constant for the apparent dephosphorylation of [18F]FDG-6-P (k4*) were significantly greater than zero and fell with increasing duration of the experimental period. Although the TH Model included no term to describe an apparent dephosphorylation of [18F]FDG-6-P, the fit of the TH Model to the time course of total tissue radioactivity was at least as good as and often better than the fit of the 4K Model in the 120-min period following the pulse of [18F]FDG. Hence, the high estimates of k4* found in PET studies of less than or equal to 120 min can be explained as the consequence of measuring radioactivity in a heterogeneous tissue and applying a model designed for a homogeneous tissue; there remains no evidence of significant dephosphorylation of [18F]FDG-6-P in this time period. Furthermore, use of the 4K Model led to an overestimation of rCMRglc; whole-brain glucose utilization calculated with the 4K Model was greater than 20% higher than values usually obtained in normal humans by the model-independent Kety-Schmidt technique. rCMRglc was accurately estimated by the TH Model and, in experimental periods sufficiently long to minimize the effects of tissue heterogeneity, also by the original 3K Model of the deoxyglucose method.