Aiming at the limitations of attribute reduction approach based on .Pawlak attribute significance and conditional entropy, an efficient attribute .reduction algorithm based on distinguish matrix and the improved attribute .significance is put forward