In this paper , key techniques of feature extraction in reverse engineering is studied. First use a method to extract the feature points from scattered points based on Gauss curvature extreme point. Use these feature points to make cloud data segmentation. Then quadric surface feature and free-form surface feature can be separated. The quadric surface feature can be fitted by quadric surface parameter equation with LSM directly. However the free-form surface feature should be extracted based on section feature recognition. This method has several steps: First ,a adaptive slicing algorithm is used to get section data from point data . Then feature points from section data is extracted. And these feature points are regarded as the gist to recognize section feature. When the section feature is recognized, they should be fitted with LSM, and then skinning method is used to reconstruct free-form surface modeling.