In order to effectively monitor the engine running state, a fault diagnosis method is put forward on the combination of the wavelet packet and neural network engine. Aimed at the petrol engine of a miniature car, the test pilot system based on vibration signal analysis is built, vibration characteristic parameters of the engine are collected in normal operating conditions and fault conditions. When the wavelet packet is decomposed and reconstructed, the eigenvectors are extracted to demonstrate the working conditions of the engine, as the neural network training sample data and test sample data. BP neural network is inputted and then trained, to implement the fault type identification of the engine with the desired effect.