For the fault recognition of rolling bearings, a fault recognition method based on Empirical Mode Decomposition (EMD), Genetic Algorithm (GA) and BP neural network is proposed. This model optimizes the initial weight and threshold by the Genetic Algorithm. Moreover, the output error of the training data is the objective function. During the process of fault recognition, the empirical mode decomposition (EMD) energy ratio as the input of the neural network is used to recognize the fault of rolling bearings under different conditions. The results of numerical simulations show that the method is better than the traditional BP neural network in the convergence precision, the recognition rate and the convergence speed.