Abstract: For the large scale scattered data，a two-step adaptive surface reconstruction approach is developed．The first step employs the adaptive extraction of key points based on the measure of curvature to produce the topologic mesh．It follows the second step by reconstructing the surface from the topologic rectangular mesh based on the NURBS method．By this approach，the large scale scattered data can be reduced to the reasonable scale，while main characteristics of the whole scattered data are remained，thereby the precise surface reconstruction with NURBS is realized．