In view of the deficiency of the present attribute weight methods based on the rough sets theory, the author proposes one new comprehensive attribute weight method through studying deeply attribute importance on the basis of rough sets theory. The proposed method considers objective weight and subjective weight. The objective weight includes three factors, named as the importance of the attribute itself, the increment of mutual information, and its own information entropy. The subjective weight is obtained by the experts with prior knowledge in the field. Experiment results prove that the new method not only overcomes the deficiency of the existing weight methods, but also is more in line with the actual situation.