To improve the accuracy and overcome the flaws of single neural network, an integrated neural classifier for stream turbine vibration fault identification is proposed based on particle swarm optimization (PSO) in the paper. The method firstly establishes diagnosis decision table of stream turbines from fault sources to fault symptoms based on wavelet package decomposition technique to faults wave-shape. Then the discrete decision table is acquired by quantifying attribute values in decision table using information entropy, a simplified decision table then is generated by rough set reduction. Based on it, several neural networks are applied to identify steam turbine faults at the same time, and their results are integrated with PSO-based. Both simulation and trial in stream turbine damage identification indicate that the proposed method has higher identification rate and shorter training time as well as excellent generalized ability, and is an ideal pattern classifier.