Mobile laser scanning (MLS), which recently has been developing so quickly as a promising technology for mapping and remote sensing (RS), offers a good means to measure the fundamental geographic data, e.g. culverts, for urban planning and road engineering. This study as the first try presents a new automatic method to detect culverts in MLS point clouds, in which actually only partial characterization of this category of objects can be presented due to the restricted scanning zenith of MLS. The schematic is based on the raster-form of the data, and the digital terrain models (DTMs) with multi-leveled resolutions are first yielded by local minimum filtering. Then, the common layout of the expanded areas containing culverts is generalized as the theoretical basis, and the schematic components are derived to deploy the concrete judgment. The geometry and intensity information about culverts are both utilized to determine the real locations from coarse- to fine-scales. Numerical analysis based on the real-measured MLS data at the Espoonlahti test site has basically validated the proposed approach. Concretely, the statistical errors of the retrieved lengths and widths of the pedestrian culverts are less than 9% and 16% compared to the real ones individually, notwithstanding the inner heights innately in-accessible.