According to the non-stationary, mutability and short duration of power quality disturbance in distribution network system, a new method to localize and identify disturbance is proposed based on lifting wavelet and improved BP neural network. At first, the Euclidean decomposition principle is used to obtain db4 wavelet lifting scheme. Then, disturbance signals are decomposed through lifting wavelet analysis, and the mutation peak of power quality disturbance is localized using lifting wavelet modulus maxim. At last, traditional BP algorithm is improved by combining increasing momentum method with self-adaption learning rate method, and improved BP neural network is used to identify disturbance signals. The simulation results show that the proposed method can better localize the disturbances’ start-stop time of distribution network system with relatively high accuracy, and can effectively overcome the shortcomings of traditional BP neural network which is easy to fall into local minimum and has slow convergence speed, and can identify power quality disturbance in distribution network system with high discrimination ratio.