The classification of power quality disturbance is important for the comprehensive evaluation of power quality and the management of disturbance source location. In this paper, a classification method for power quality disturbance based on generalized S transform and Differential Evolution Extreme Learning Machine is proposed. Firstly, window width factor of different frequency area in the S transform is changed to improve the performance of features. Secondly, Differential Evolution Extreme Learning Machine is used as the disturbance classifier, which adopting the differential evolution algorithm with global optimization. The Differential Evolution Extreme Learning Machine optimize the input weight and bias of hidden layer nodes to enhance the generalization ability of Extreme Learning Machine, and further improve the classification accuracy. As simulation experiments shown, comparing with the Support Vector Machine and Extreme Learning Machine, the new method can get higher accuracy and stronger noise resistance, so it is suitable for the identification of power quality disturbances.