In view of non-linear characteristics between fault symptoms and fault types of transformers, a wavelet neural network fault diagnosis model based on improved particle swarm algorithm is designed with the data of dissolved gas analysis. The model, constructed by three-layer wavelet neural networks, is trained by an improved particle swarm algorithm. By introducing the mutation operator of genetic algorithm, inertia weight factor and Gaussian-weighted global extremes on the basis of the standard particle swarm algorithm, it can accelerate the training speed of wavelet neural network and improve the accuracy of training. The simulation experiments show that this improved particle swarm algorithm wavelet neural network can be effectively applied to transformer fault diagnosis and provides a new way for transformer fault diagnosis.