A classification method for short duration power quality disturbance based on S -transform with module time-frequency matrixes partial similarity is proposed. At first， all types of the disturbance characteristics with different duration are standardized by time-frequency scale zooming. The standard templates of various disturbances with difference size matrixes and obvious characteristics are built according to the special part of module time-frequency matrixes of disturbance characteristics. Then， comparing disturbance signal with standard template， the similarity can be calculated， and disturbances are classified according to maximum similarity principle. Due to various disturbances in different time-frequency area have a high polymerization degree in their characteristics， application of partial similarity theory in time-frequency area has greatly enhanced the similarity of same type disturbances， obvious difference among the standardized templates of different disturbances is built. So it can achieve the disturbance classification directly without using auxiliary classifier. Besides， it's clear and simple， suitable for disturbances with different duration. Because the clear time-frequency relation in local area is effectively used， the method has stronger anti-jamming capability than that of overall similarity. The simulation results show that it has a good effect for classification of short duration power quality disturbance.