Terrestrial laser scanning (TLS) point cloud data were gathered in the field of exposed gravel surfaces for sampling sites with a spatial extent of 6m × 6m. We propose in this letter a novel two-stage mean-based filter scheme for processing the point cloud to generate the digital surface model (DSM) at a resolution of 1 cm. To the authors' knowledge, this letter reports the first work of generating a DSM of a gravel surface of such a large spatial extent with high spatial resolution using TLS. The elevation variations attributed to gravel clasts and single grains can both be captured in the point cloud data. To eliminate data voids due to the obstruction of the line of sight, a multiple-scan strategy is employed, which includes four scans at the corners of the sampling site and two supplementary scans for the central 2m × 2m area. The resultant DSM exhibits good agreement with elevation profiles obtained using a traditional manual profiler. The proposed method is an effective tool for obtaining a quality DSM of fluvial gravel surfaces using TLS with fewer scans than previous study.