An automated method was proposed for anatomic standardization of PET scans in three dimensions, which enabled objective intersubject and cross-group comparisons of functional brain images. The method involved linear scaling to correct an individual brain size and nonlinear warping to minimize regional anatomic variations among subjects. In the linear-scaling step, the anteroposterior length and width of the brain were measured on the PET images, and the brain height was estimated by a contour-matching procedure using the midsagittal plane. In the nonlinear warping step, individual gray matter locations were matched with those of a standard brain by maximizing correlation coefficients of regional profile curves determined between predefined stretching centers (predominantly in white matter) and the gray matter landmarks. The accuracy of the brain height estimation was compared with skull x-ray estimations, showing comparable accuracy and better reproducibility. Linear-scaling and nonlinear warping methods were validated using [18F]fluorodeoxyglucose and [15O]water images. Regional anatomic variability on the glucose images was reduced markedly. The statistical significance of activation foci in paired water images was improved in both vibratory and visual activation paradigms. A group versus group comparison following the proposed anatomic standardization revealed highly significant glucose metabolic alterations in the brains of patients with Alzheimer's disease compared with those of a normal control group. These results suggested that the method is well suited to both research and clinical settings and can facilitate pixel-by-pixel comparisons of PET images.