In order to obtain the boundary feature of unorganized points, a method of boundary feature recognition and abstraction was proposed based on the kernel density estimation on k-neighborhood of every sample point. k-neighborhood of a sample point could be acquired quickly based on R*-tree index, and the radius of query area viewed as bandwidth was used to kernel density estimation on the point set consisted of the sample point and its k-neighborhood. In this way, the mode points reflected the distribution of point sets could be obtained. According to the ratio of distance between mode points and sample points to the bandwidh of kernel density estimation, the sample points located on boundary could be recognited and abstracted. The experimental results show that the algorithm can obtain the boundary feature of the unorganized points in uniform or nonuniform distribution exactly and rapidly. 