To overcome the difficulty of automatic identification of power quality disturbances from the large data of power quality monitoring system, a new method for power quality disturbances identification is proposed based on generalized S-transform and PSO-PNN. It makes full use of generalized S-transform’s ability of giving attention to both time and frequency resolution. Initially, the time-frequency analysis of power quality disturbances is carried out by using generalized S-transform, from whose results the time-frequency features of disturbances are extracted. Finally, PSO-PNN, as a classifier, is used to identify power quality disturbances. The PSO algorithm solves the problem of choosing the smoothing factor for PNN which is usually hard to determine, and thus the performance of the classifier is greatly improved. The simulation results show that the proposed method can identify six kinds of power quality disturbances correctly and effectively, and it is characterized by high recognition correctness rate and low sensitivity to noises, and it will find extensive application. This work is supported by National Natural Science Foundation of China (No. 61561007).