Abstract: 3D point cloud data have received great attention, and feature detection of the unordered and the large mount point data is hot topic for the recent years. We presented a feature point detection algorithm based curvature and density. Firstly, feature parameter of each point is calculated. The parameter includes three parts: the average distance of the neighboring points, the sum of the normal angle between the point and its neighboring points, and the data point curvature. Secondly, we define the density of data calculated by using Octree, which is applied as the feature threshold to determine feature points. The feature point is recognized when its density parameter is bigger than the threshold. In this article, we only use the geometry properties, such as normal of point, curvature and the neighboring points to detect the feature points. The experimental results show that our new approach can accurately detect the feature poinst for 3D scattered point data cloud models.