Abstract: Remote sensing data is widely used in earth resources investigation and has made a great progress. Unfortunately,because the computer classfication is based on multispectral features of images,the accuracy and realiablity are limitted. To improve the accuracy and realisblity of recognizable classes of sround objects,many people have been striving for a long time and many methods have been prompted. A better way is to simultaneously use many kinds of auxiliary data, such as DTM,geographic features soil types, climate, relief, vertical and regional distribution etc.,with remote sensing image in computer classification. For human interpreter,it is not a very difficult problem. But it is not easy for computer. At the moment there is a tendency to develop Expert System in remote sensing image computer classification. It is an efficient way to solve this problem.The structure of Expert System that uses TM image for gtass resources classification is described. Two ports are included in GES:1)Hish level processing part. This port includes the following modules. Knowledge base, Inference engine, Database, Nature language module, Explain module, Knowledge acquisition module etc.. For inference engine,grey inference theory is introduced to inexacting inference.2)Low level processing part. This part is designed to extract information from image and auxiliary data. Such as DTM,soil types, texture features etc.. The image segmantution and output of classification results are also completed by this part.