Difficulties in obtaining satisfied sample data of remote sensing image and the lack of effective methods to accumulate and manage them are the bottlenecks for the development of classification technology. Against these problems

a CBR(case based reasoning) system for remote sensing classification is established on the base of modified Fuzzy ARTMAP neural network in this paper. In the CBR system

the modified Fuzzy ARTMAP neural network plays the roles of the knowledge extractor of cases and the classifier of remote sensing images. Reasonable reserve

optimization combination and reutilization of remote sensing samples can be implemented with the CBR strategy. The TM image of Xiang Hai nature reserve is classified by following four methods respectively: the CBR system

maximum likelihood

BPNN and modified Fuzzy ARTMAP. Experimental results show that

comparing with other classification methods

both the utilization efficiency of remote sensing samples and the classification accuracy of images can be greatly improved by the CBR system. The problem of how to utilize the existing samples efficiently for remote sensing classification when the sample data of the current image are limited can be solved to a certain extent in the proposed system.