Parts of two multispectral Landsat scenes, compressed by ratio 6·4/1, were investigated on the effect of the compression on the image contents. Landsat images of the Bayuda desert in Sudan and the Baltic Sea have been submitted to supervised maximum likelihood classification. Partitioning of classified images is given on a percentage basis as a function of sigma. Three-dimensional clusters of spectral channels and histograms show the influence of compression. Discussion of results shows effect of compression on partitioning of classified scenes, being displayed in several tables and figures.