A procedure is described for automatically separating temperate fruit tree orchardsfrom mixed deciduous forest in LANDSAT Thematic Mapper data. A filter is applied to enhance the texture of bands 4 and 3 prior to ratioing, smoothing, and level-slicing to a binary image on the basis of supervised training. Including the binary image in supervised classification of single date imagery has reduced misclassification of forest as orchard from 75 percent of the forest pixels to fewer than 7 per cent.