A fast raster rotation algorithm based on nearest-neighbour interpolation is described. Essentially, the computation consists of two additions/subtractions and two rounding operations per rotated image pixel. The rotation time is about equal to the time needed for the pixelwise duplification of an image of the same size. 'Nearest-neighbour' interpolation and 'ideal' interpolation are compared. Because of non-linearities introduced by binary threshold operations, both interpolation functions are found to be equivalent. It is shown that image frequencies should not be higher than one quarter of the sampling rate. In order to rotate large images, an image memory management module is used for swapping transparently image blocks between main memory and disk.