Abstract: This paper presents design principle and realizable approach for Inference Decider for Recognition of Soil Classification Remote Sensing(IDRS) in arid land. On the basis of non-supervising classifying for soil classifications with TM remote sensing images, the author builds reasoning mechanism of direct inference combined with contrary reasoning for soil classification and recognition decision. The author also expresses soil classification and recognition knowledge of expert in soil with data structure of producing rule linking with frame rule for knowledge expression in the IDRS. Furthermore, the author structures rules of soil classification and recognition with image structure model, and builds decision tree of soil classification in the IDRS, and organizes file of decision for soil classifications with typical image case model. With the means, the author makes a test research on classification and distinguishing for soil in test region of Fukang County, situated on the northern foot of the Tianshan Mountains, Xinjiang Region. And the test result shows that the approach mentioned above is a high reliable precision, and it reclaims a new way for classification and recognition of soil in arid land.