In order to identify the disturbance signal in the power system and reduce the influence of disturbance signal on system security,a voltage disturbance signal classifier based on improved local mean decomposition and BP neural network is proposed. Disturbance signals are decomposed in three layers by LMD,obtaining product function components with amplitude and frequency information of the voltage signal. The signal energy value constructed by the PF components is as an input of the BP neural network to identify and classify the voltage disturbance signal. Experiments on the four typical voltage disturbance signals show that the signal classifier based on LMD and BP neural network has high accuracy and good working efficiency in voltage disturbance signals recognition and classification.