In order to improve the ability of complex power quality disturbances recognition, a new type of complex disturbances recognition approach based on Multiresolution Generalized S-transform (MGST) is proposed. Firstly, the spectrum of original signals is segmented into 3 frequency areas including low frequency area, medium frequency area and high frequency area. The width factor of window function in S-transform is defined respectively in different frequency areas. MGST has different time-frequency resolution in each frequency area in order to satisfy the recognition requirements of different disturbances in each frequency area. Otherwise, the width factor of window function in the high frequency area is adaptively adjusted according to the value of Fourier spectrum of the fundamental frequency. On this basis, the decision tree based on 6 features is constructed to recognize disturbance signals. Finally, the minimum classification faults rule is designed to get the optimum threshold of each node. The simulation and real signals experiments show that 13 types of disturbances including 5 types of complex disturbances are recognized accurately by the new approach. The new approach has better classification accuracy and noise immunity than other methods such as S-transform , generalized S-transform and Hyperbolic S-transform.