Misfiring fault is one of the common faults of automobile engines. This paper presents an algorithm based improved neural network which is used for misfiring fault. It calculates the memberships of inputs and initializes the weights and thresholds of the neural network by genetic algorithm firstly, and then trains the improved neural network and uses it for diagnosis. By applying GUI function of MATLAB, a new man-machine interaction interface was designed. The results of experiment indicate that this algorithm can effectively carry out misfiring fault diagnosis.