A new method based on neural network and genetic algorithm to optimizate the Multiphase Rotodynamic pump is given. Using cubic B-spline surface to parametric the blade profile. Based on the ability of highly nonlinear fitting of BP neural network, the nonlinear relation between the blade parameter and the pump performance parameters is build. Let the trained neural network as a fitness function of the genetic algorithm, using the characteristic of nonlinear global optimization of genetic algorithms to optimize multiphase rotodynamic pump. Through the fluent numerical calculation of the genetic algorithms output value, the results show that the capability of multiphase pump blade is improved, and then proved the feasibility of the optimization method.