Regarding weapon system cost estimation the paper aims at three key problems: the establishment of parametric models,the establishment of objective function and the determination of optimum parameters.The related analysis is extended respectively.Considering the adaptability of models to samples,radial basis function neural network is adopted to establish the parametric model;considering the inconsistency between fitting precision and generalization of models,an eclectic method is adopted to establish the objective function;and considering the multiple parameters and the complex format of objective function,particle swarm optimization algorithm is adopted to optimize model parameters.A verification example is provided.Theoretical research and numerical analysis show that the method can thoroughly solve the problem of model establishment and parameter optimization,supply a theoretical guidance to the new parametric model and has a good performance of currency and popularization.