A new method of data mining based on rough set and neural network is proposed. Based on the rough set theory, attribute reduction is processed on data under the consistent conditions. Then neural network is used to study and predict data, as well as reduce the attributes under the inconsistent conditions. Finally rule knowledge in the neural network is extracted by using rough set theory. The method mixes rough set's strong attribute reduction, rule extraction ability and neural networks classification, robustness ability. Experimental results show that this algorithm can produce more effective and simpler rules quickly and possess good robustness.