A free form surface reconstruction method based on least square support vector regression is presented. Firstly in order to eliminate noise points, some sample points are chosen from the measured data to construct LS-SVM model. Thus a LS-SVM model to approximate the measured points is obtained. And the distribution probability of the approximation error is figured out. In result, the noise points are eliminated when their error probability is less than the specified threshold value. Then the boundary points are extracted. Lastly the surface model is reconstructed by use of the measured points from which noise points have been eliminated. The results indicate that the reconstruction precision can satisfy the demands of engineering application.