For the distribution characteristics in a slice of pathological cell image, the system transforms them into the characteristic vector by quantization and clustering in HSV color model, it promotes the concerned isolated pixel color description into the color feature of its neighborhood region color histogram. By choosing appropriate neighborhood window size, it uses color information entropy method to gain efficient primary classification of gastric slice image and the similar retrieval results from the sample library images. Experiments show that this method has a better image classification and retrieval result.