Abstract: The artificial neural networks are applied to the prediction of the specific impulse for solid propellant rocket engine. This method avoids the difficulties of concrete law analysis and the mathematical modeling. We can obtain directly the network model which contains the relation of actual system. An improved radial basis function neural networks is presented, which compensates the defect of undiscovered number of radial basis function for the traditional radial basis function neural networks. The forecast results of radial basis function neural networks and back propagation learning algorithm are compared.