An image can be sampled by taking as sample values either the image intensity at defined sample points or the image intensity integrated over regularly spaced raster cells . The latter concept, termed integral sampling, is investigated using deterministic and statistic methods . It is proved that the use of integral rather than point sampling results in reduced error due to aliasing when an image is reconstructed from its samples . Filters for optimal reconstruction from samples taken in the presence of noise are described . The special cases of integral sampling of high-pass filtered images and of binary pulse-code modulated images are analysed . Integral sampling can be performed optically ; it is shown to be of practical interest for image evaluation (thematic mapping) and for optical spatial regenerative repeaters .