SummaryTwo modifications to the conventional procedure of crosscorrelation are described, widely used for establishing the relative alignment of the members of a set of images from which a higher resolution or more interpretable restoration is sought. Both achieve a high and sharp peak in circumstances where the conventional peak is too ill defined to be recognizable: neither involves significant additional computation time. The more general method requires rough knowledge of the imaging conditions, but a variant applicable to images with axial resolution has no such requirement. In addition, a least-squares procedure is presented for achieving an optimum compromise between many pair-wise displacement measurements, preventing the accumulation of alignment errors across a set of images.