Unsupervised and supervised classification of soils of an arid area in Xingjiang was made using landsat data which were derived from scanning of the images on a transparent negative hard-copy and processed by micro-computer.The map compiled from these data is on a scale of 1:400,000 and covers an area of 12,375 km².Before classification,the scanned data of the four channels were matched by columns and rows according to some special topographic fertures on the image.All anomalous data were rectified by the methods designed by the author.For the unsupervised classification,the key tactics adopted by the author was firstly to find out and fixed the suitable number for the division of soil groups in this region, and then to select the proper parameters in order that the iteration could be continued freely until satisfied results were obtained.