To effectively identify the bearing fault position and the degree of damage,a model of adaptive cuckoo search algorithm combined with BP neural network（ ACSBP） is proposed. In the ACS algorithm,the Levy flight strategy is eliminated to reduce the randomness of the search process. Simultaneously,in order to reflect the universality for different optimization problems,the updating of step size is implemented by the fitness function value,which did not need to be initialized. Furthermore,the dynamic adjustment method of discovery probability is adopted to improve the optimization accuracy and convergence rate. Based on these approaches,the model of bearing fault diagnosis is established. The diagnostic results show that the ACSBP algorithm has stronger fault tolerance compared with CSBP and PSOBP models,and can effectively improve the accuracy of bearing fault diagnosis.