The large number of raw point cloud data collected with three-dimensional laser scanners presents a challenge during the subsequent data processing. Unfortunately, the existing curvature-based point cloud compression methods can lead to loss of details in the sub-feature regions. Therefore, we propose a curvature-grading-based compression method for point cloud data in this study. First, the feature distribution is obtained by estimating the curvature of every point. Then, the curvature level of each point is acquired based on the logarithmic function and its normalized curvature. Finally, voxelized grids are created over the input point cloud and are used to perform grading compression according to the levels. The experimental results denote that the proposed method can preserve the details of raw data while reducing the amount of data, resulting in an efficient pathway to compress the point cloud data.