Abstract: This paper describes a new method for constructing a rough neural network. In this network, rough neurons lie in the hidden-layer, and they consist of three parts which generated by two hyperplanes that partition the universe. The hyperplanes are obtained by a method that is similar to support vector machine (SVM). This neural network has the characteristics of definite configuration, good understandability, simple computation and fast convergence. An example based on the aircraft actuator failure classification is presented. Simulation results show that this method is effective.