To improve the registration accuracy of point cloud data generated by 3D laser scanning, a new method of iterative interpolation registration based on fast point feature histograms (FPFH) was proposed. Due to the effect of the scanner's resolution in the process of registration, partial or overall density of obtained point cloud data was smaller so that there were no same points even the measuring locations of point cloud data were fixed. Therefore, errors existed between corresponding points. In order to reduce the influence of these errors on the registration accuracy, iterative interpolation method was introduced to increase the overall density of point cloud. The FPFH features of key points were used to find the corresponding relationship; random sample consensus algorithm was used to remove the false correspondence between two point clouds; then the coarse registration rotation and translation matrix was gotten by using singular value decomposition algorithm on the corresponding covariance matrix; at last, the iterative closest point algorithm was employed for the precise registration of point clouds. The experimental results show that the improved registration algorithm is simple, stable and reliable and its computation velocity is faster. This method effectively improves the accuracy of registration results.