Abstract: A method based on the classical rough set(CRS)for determining the case feature weights is introduced and improved in order to overcome problems of the existing algorithms such as poor objectivity and high complexity.For the problem that the accuracy of case-based reasoning(CBR)degrades due to measurement errors of the case similarity caused by the CRS methods,a new method based on similarity rough set(SRS)for determining the case feature weights is proposed where the indiscernibility relation is extended to the similarity relation of the CRS.The definitions of SRS and two theorems of computing the feature weights based on discernibility matrix are presented.And the proposed methods are proved to be effective with a practical example.