Abstract: It is difficult to understand the actual flow inside a centrifugal pump, whose design is still semi-empirical and semi-theoretical and whose prediction accuracy is not satisfactory. Thus, we establish a new model for predicting its performance. We combine the genetic algorithm with the BP neural network to conduct the optimal design of its performance parameters. We use the genetic algorithm to optimize the weighted values and threshold values of the BP neural network, and then use the BP algorithm to train the neural network, thus avoiding the local minimum values when the training is done with the BP neural network alone. We use some representative performance data of the centrifugal pump as training samples of the new model. The study results show that the BP neural network optimized with the genetic algorithm speeds up the convergence of the model and improves the training accuracy. Our method can serve as an effective theoretical basis for the optimal design of performance parameters of a centrifugal pump.