To improve the accuracy of Radial Basis Function (RBF) approximation model, the influencing factors on approximation accuracy were deeply studied. The truth that matrix condition number and shape parameter were two important factors of approximation accuracy was pointed out by analyzing the influence of rounding error over approximation accuracy thoroughly. The matrix condition number was decreased and the design freedom was increased by separating design space based on sensitivity analysis. Learning from the traditional RBF based on optimal shape parameter, the construction method of RBF approximation model based on parameter optimization of space decomposition was proposed. The numerical test results show that, in the two cases, the Root Mean Square Error (RMSE) of the proposed method is reduced by 51.3% and 58.0% respectively while comparing with the traditional method based on optimal shape parameter for construction of RBF approximation model. The proposed method has high approximate accuracy.