Accurate Digital Terrain Models (DTMs) are inevitable inputs for mapping and analyzing areas subject to natural hazards. Topographic airborne laser scanning has become an established technique to characterize the Earth's surface: lidar provides 3D point clouds allowing for a fine reconstruction of the topography while preserving high frequencies of the relief. For flood hazard modeling, the key step, before going onto terrain modeling, is the discrimination of land and water areas within the delivered point clouds. Therefore, instantaneous shorelines, river banks, and inland waters can be extracted as a basis for more reliable DTM generation. This paper presents an automatic, efficient, and versatile workflow for land/water classification of airborne topographic lidar points, effective at large scales (> 300 km2). For that purpose, the Support Vector Machine (SVM) method is used as a classification framework and it is embedded in a workflow designed for our specific goal. First, a restricted but carefully designed set of features, based only on 3D lidar point coordinates and flightline information, is defined as classifier input. Then, the SVM learning step is performed on small but well-targeted areas thanks to a semi-automatic region growing strategy. Finally, label probability output by SVM is merged with contextual knowledge during a probabilistic relaxation step in order to remove pixel-wise misclassification. Results show that a survey of hundreds of millions of points are labeled with high accuracy (> 95% in most cases for coastal areas, and > 90% for rivers) and that small natural and anthropic features of interest are still well classified even though we work at low point densities (0.5–4 pts/m2). We also noticed that it may fail in water-logged areas. Nevertheless, our approach remains valid for regional and national mapping purposes, coasts and rivers, and provides a strong basis for further discrimination of land-cover classes and coastal habitats.