Bronze mirrors have very important research values on archaeology

and the similarity evaluation of bronze mirror images is the key to solving the problems in manual classification and realizing the large scale quantitative analysis of bronze mirror images. This paper discussed the feasibility of KL expansion for similarity evaluation of mirror images

specified the normalization of mirror images and the procedure of eigen space construction through KL expansion

and proposed 3 similarity evaluation standards based on the " class" conception. Experimental comparisons and analyses of these evaluation standards show that

the 3 proposed evaluation standards based on the "class" conception are more effective than the conventional distance standard

and their evaluation accuracy reached higher than 85% under eigen space of higher dimension.