Abstract: When constructing an approximation surface with Response Surface Method（RSM），the approximation surface usually covers the whole interpolating space.Because of the cost of computation，the order of the approximation surface usually is below four.However，when constructing an approximation surface for a nonlinear surface in large space with low order response surface，the accuracy of approximation will be absolutely very low.Actually，it’s easily concluded from the Taylor theory that the response surface can only have good accuracy in small space.The Improved RSM（IRSM） raised in this paper divides the interpolating space into several small subspaces，which will reduce the scale of the space that the RSM has to cover，and improves the accuracy of RSM.The result proves that the accuracy of IRSM is never below the RSM no matter what the number of sample point is.If the number of sample points is large，the approximating accuracy of IRSM is much higher than the RSM，which makes the IRSM is more suitable for the MDO.