The generation of boundaries from the triangular meshes of CAD object is very important for successful surface segmentation and subdivision. Considering the characteristics of non-uniformity, sparseness and known facet normal for STL model, which is generated by a CAD system, a new and efficient approach for extracting boundaries on STL model is presented in this paper. Boundaries are classified into sharp feature edge and non-sharp feature edge embedded in the filleting and transitive region. The former is detected based on dihedral edge angle (angle between facets), while the latter is identified by analyzing intrinsic region properties of triangular facets distribution without curvature estimation. By means of unidirectional increment of feature edges, continuous region boundaries are generated. Successful examples are presented to illustrate the feasibility of the proposed algorithm.