Abstract: Based on a measurable survey pattern corresponding to a rocket engine failure pattern,the multi failure is detected by using back propagation (BP) neural network,which provides interesting means for pattern recognition and classification.The BP algorithm minimizes the mean square error between the desired and the actual output of the network.The learning algorithm is carried out with a gradient descent techniuqe.In order that the mean square error does not fall into partial minima noise,a sequence of random noises is recommended in the BP algorithm.As a test case,only the pump inefficiency and the sprayer jam in the liquid rocket engine were studied,and the failure detection simulation was given to show great advantages of the BP neural network.