<p>Recently, the underwater digital terrain model (UDTM) has been extensively used for seafloor terrain measurement. Identifying terrain and nonterrain points from an underwater bathymetric point cloud has been a fundamental problem in UDTM modeling. Hence, we propose a new ground filtering method derived from smoothed particle hydrodynamics to distinguish bathymetric points. The proposed method is called smoothed particle simulation filtering (SPSF), which embodies filtering as a matching of a smoothed particle grid to the point-set surface. First, four seafloor terrain features, namely curvature-based shape factor, slope, roughness, and concave–convex coefficient, are extracted from the moving least-square fitting surface of the original point-set surface, and the terrain complexity of each point is computed. Then, an enhanced particle motion model appropriate for ground filtering is introduced. To improve filtering performance, we divide the point-set surface into several subregions and set adaptive parameters for diverse subregions according to the terrain complexity. Finally, subregional point cloud filtering is performed, and each bathymetric point is marked as “terrain” or “nonterrain”. To assess the filtering performance, SPSF is applied to process the multibeam bathymetric point cloud data collected around Chi Island in Qingdao, China. Results indicate that compared with the same type of filtering methods, SPSF improves the filtering accuracy by more than 50% and the filtering efficiency by more than 40%. Furthermore, compared with other types of filtering methods, SPSF displays great potential in alleviating over- and under-filtering in seafloor terrains.</p>