Current card is the main basis of electric submersible pump well fault diagnosis. At present, the diagnosis of electric pumping wells is mainly completed by technical workers. It is difficult to achieve rapid mass diagnosis, and the diagnosis results are greatly influenced by engineer technology level. Therefore the application of BP neural network for fault diagnosis of electric submersible pump is put forward. First of all, different pump current cards are collected, and sample library is established. The eigenvalues of the different pump current card are gathered and trained according to certain principles. After completion of training a weight matrix is got, and the current card characteristic values and weights matrix of similarity is calculated. Through computer programming application, the method can be proved to make electric pump diagnosis accurately and rapidly.