Abstract: Traditionally, technical engineers often rely on their experience to manually diagnose ESP wells failure according to electrical current charts which is the main evidence for ESP wells diagnose. This makes it hard to achieve batch diagnosis, and the diagnosing accuracy is significantly affected by the technical level and working state. BP neural network technology applied to diagnosing ESP wells failure is addressed in this paper. First, characteristics of electrical current charts under different conditions has been analyzed to obtain the characteristic sample by integrated experts experiences; second, the necessary weights and threshold values for neural network calculation are given by learning and training the sample set; and then been saved; third, the character data of electrical current cards which need to be analysis is putting to calculate the similarity by using the BP neural network identification method; finally, the failure type of ESP is determined by comparing the calculated similarity to the most nearly sample.The computer program of this technology is used on site, and application results has proved that the method is reliable and robust, and can be utilized to batch diagnose ESP well failures correctly and quickly.