A discernibility matrix is a good idea for attribute reduction. Discernibility information tree can effectively eliminate redundancy elements in discernibility matrix, and it realizes the compactness storage of discernibility matrix. However, discernibility information tree neither considers the role of core attribute in eliminating redundancy elements nor considers the effect of attribute order in the compactness storage elements in the discernibility matrix. Therefore, an improved discernibility information tree is proposed, which can further compress the storage space of non-empty elements in the discernibility matrix. In order to verify the effectiveness of the improved discernibility information tree, some simulation experiments for UCI datasets are displayed.