We have developed, validated, and employed a technique of retrospective spatial alignment and integrated display of positron emission tomographic (PET) and high-resolution magnetic resonance (MR) brain images. The method was designed to improve the anatomical evaluation of functional images obtained from single subjects. In the first computational step, alignment of PET and MR data sets is achieved by iteratively matching in three orthogonal views the outermost scalp contours derived from front-to-back projections of each data set. This procedure avoids true three-dimensional modeling, runs without user interaction, and tolerates missing parts of the head circumference in the image volume, as usually the case with PET. Thereafter, high-resolution MR sections corresponding to the PET slices are reconstructed from the spatially transformed MR data. In a phantom study of this method, PET/MR alignment of the phantom's surface was accurate with average residual misfits of 2.17 to 2.32 mm as determined in three orthogonal planes. In-plane alignment of the phantom's insertion holes was accurate with an average residual misfit of 2.30 mm. In vivo application in six subjects allowed the individual anatomical localization of regional CBF (rCBF) responses obtained during unilateral manual exploration. In each subject, the maxima of the rCBF activations in the hand area were precisely allocated to gray matter in the anterior or posterior wall of the central sulcus. The configuration of the rCBF responses closely followed the gyral structures. The technique provided a better topographical understanding of rCBF changes in subtraction images of PET activation studies. It opens the perspective for studies of structural-functional relationships in individual subjects.