Extracting boundary curves of holes from point based models is one of the basic and key technologies in the research field of digital geometry processing. An algorithm was presented for extracting the boundary curves based on the properties of point clouds and deriving a combined criterion for automatic hole detection. First, the k-neighboring points for each point were collected in a given point-based model and then the normal of the point-set surface at the point is evaluated. For each point, several essential criteria were combined, such as the largest angle criterion, half-disc criterion and shape criterion respectively, into a weighted blended probability. Finally, the boundary curves of holes were constructed by searching the nearest point strategy and the boundary points were connected by a poly-line to generate the boundary curves. Experiments show that this algorithm can run stably and robustly, and can extract various hole boundary curves from point cloud models.