Aiming at the classification and recognition problem of composite power quality disturbances, a composite power quality disturbance recognition algorithm based on piecewise-modified S transform and RBF neural network is proposed. Firstly, the S transform is modified by segmenting the time resolution and the frequency resolution. By analyzing the obtained mode time-frequency matrix, the characteristic curve that can reflect different mutation parameters of the disturbance signal is drawn. Secondly, 10 types of characteristic parameters for pattern recognition are extracted by using statistical methods and optimization. The RBF neural network classifier is designed to classify the extracted feature samples by training and classification. Finally, six types of composite power quality disturbance classification including five single disturbances and harmonic and voltage sag, voltage sag and flicker, etc. are simulated under different noise environment. The simulation results show that the proposed scheme is superior to S transform and global approximation networks in terms of time-frequency processing ability, classification ability and learning speed, and is robust and can accurately identify multiple kinds of single disturbances and two kinds of disturbances simultaneously composite power quality disturbance. This work is supported by National Natural Science Foundation of China (No. 61863023).