A rapid image matching algorithm for cerebral three-dimensional (3D) images is described. Fully automatic 3D image matching between images acquired the same modality and this was realized by applying the following sequential processes: (1) the calculation of one-dimensional (1D) projection patterns from both reference and input 3D images; (2) the matching between the two sets of 1D patterns; and (3) the transformation of input 3D image according to the pattern matching result. The 3D morphological variations among images due to shift and linear/non-linear scaling along projection axes were recovered, as well as compensation for mismatching due to image rotation and partial deficit of image data. The computation time of this method was quite short compared to that of the conventional 3D pattern matching method. (C) 1998 Elsevier Science Ireland Ltd. All rights reserved.