Research on 3D face models relies on extraction of feature points for segmentation, registration, or recognition. Robust feature point extraction from pure geometric surface data is still a challenging issue. In this project, we attempt to automatically extract feature points from 3D range face models without texture information. Human facial surface is overall convex in shape and a majority of the feature points are contained in concave regions within this generally convex structure. These “feature-rich” regions occupy a relatively small portion of the entire face surface area. We propose a novel approach that looks for features only in regions with a high density of concave points and ignores all convex regions. We apply an iso-geodesic stripe approach to limit the search region, and apply the shape-index integral projection to locate the features of interest. Finally, eight individual features (i.e., inner corners of eye, outer corners of eye, nose sides, and outer lip corners) are detected on 3D range models. The algorithm is evaluated on publicly available 3D databases and achieved over 90% accuracy on average.