We present a simple, automatic method for extracting feature curves, called crest lines, from point clouds. Crest lines are surface shape features having a mathematical background. Given an unstructured point cloud as input, we pre-process the data to generate some topological information by creating an undirectional surface graph. The method starts with the approximation of normal curvature on every point. Utilizing the crest point definition, the crest points are identified. Region growth follows to implicitly connect the crest points and create crest graphs. Finally, the computation of minimum spanning trees for every crest graph and the pruning of short branches create crest lines.