Abstract: A new double-layer learning method of rough set is put forward,which can carry out the outer layer learning by genetic algorithms and the inner layer learning by rule extraction.Firstly,the genetic algorithm is introduced,the attributes are coded into binary codes,and the rules are extracted under various attribute combinations.Secondly,the test samples are employed to test the obtained rules,and the fitness function is constructed on the basis of the obtained recognition rate.Finally,the optimum attribute combinations and the corresponding knowledge rules are obtained with proper genetic operators.In comparison with other methods,the presented method combines attribute reduction with rule extraction,and possesses a stronger self-adaptive ability.In the end,examples are given to verify the proposed method.