This paper presents a method for 3D point cloud denoising using super-voxel structures. First, we define a topological relation and a distance metric of points based on the local coordinate system. Then, a K-means method is used to generate over-segmentation results, e.g. super-voxels. The super-voxel structure provides a means to produce an elastic framework over the original data. An iterative denoising method is then conducted to project points to an expected surface. Finally, we use a normal consolidation process to further adjust the point distribution. The efficiency of the proposed method is verified by experiments and comparisons.