To improve the reliability of rocket gun and make managers carry out fault prediction and prior maintenance, a fault prediction method based on adaptive fuzzy H network is proposed. A general fault prediction model for each system of rocket gun is established according to the working principle of each system. Using the historical data of rocket gun faults, adaptive learning is done by BP algorithm which can adjust the learning rate automatically, and the fault prediction of electrical system in rocket gun is simulated and analyzed. The experimental results show that this method can obtain accurate fault prediction results and achieve the purpose of rocket gun fault prediction.