This paper proposes a fault diagnosis method for analog circuits based on a combination of wavelet packet transformation and SOFM network, which uses wavelet packet decomposition as pre-processing and denoising tool for multi-scale decomposition and denoising on the sampling signal of electric circuit, and for energy feature extraction and normalization as sample input to the neural network. Neurons of SOFM network competitive layer are used in the fault identification and classification of the sampling data, which overcomes the disadvantage of difficult selection of BP network hidden neurons. This article presents a detailed description of the fault diagnosis principle and procedure, and examples of fault diagnosis are given.