Electromechanical actuator has been used in the control of flight control surface of civil aircrafts. As a key component of electromechanical actuator system, the correctness of the sensor output has a great influence on the operation of the system. Therefore, fast and effective fault detection for sensor is very necessary. The sensor types and failure modes of electromechanical actuation system were analyzed. Aiming at the current problems that the fault detection needs a lot information and long detection time to deal with liquid and nonliquid faults, twolevel GA-BP neural network was designed to diagnose and identify sensor faults, and different training methods were compared. It was determined that the fault diagnosis classification results of neural network using Levenberg-Marquardt learning method are more accurate, and then the neural network was optimized by genetic algorithm. Finally, the effectiveness of the method is verified by the experimental platform of the electromechanical actuation system. The innovation lies in the use of a twolevel network architecture, by which liquid faults and nonliquid faults can be detected rapidly, so the amount of information required for fault detection and the detection time are reduced effectively.