In view of the importance of surface shape fitting precision of free-form surface model in researches such as free-form surface representation and selection of initiating structure of surface shape, the influence of different sampling point distribution patterns on surface shape fitting precision of the Gaussian Radial Basis Function Based on Slope( Gaussian RBF-BS) model was investigated. An off-axis quadric surface and an bumpy paraboloid were fitted with different sampling point distributions which were uniformly random points and edged-clustered points. The results show that the uniformly random grid is better for surface shape fitting due to the high precision. The influence of sampling point quantity on fitting precision remains constant when certain fitting precision is reached. The off-axis Three Mirror Anastigmat(TMA) system was optimized experimentally by using initial surface shapes generated with different sampling methods. The results show that, by using initial surface shape generated from uniformly random sampling method for the system optimization, the average Modulation Transfer Function (MTF) of whole field of view is more than 0.72, which is much higher than that by using surface shape generated by edge cluster sampling points. Thus the theoretical research results were verified.