This paper presents an original algorithm (ExtracTIN) for automatic extractions of hydrographic networks using elevation data. It is applied to Mediterranean watersheds subjected to flash-floods events in the Cevennes region. The goal is to analyse the role of watersheds and hydrographic networks morphologies in upslope catchments in their hydrological response when subjected to heavy rainfalls events. The proposed method combines an analysis of networks morphologies and surface flow directions to extract what could potentially be the most extended network during rainfall events: the Network of Reconnected Thalwegs (NRT). Working on a Triangulated Irregular Network (TIN) scheme, the extracted networks are compared to those obtained with some classical raster methods and with the known hydrographic networks reported in IGN© database. All these networks are finally evaluated by comparison with a network digitalization on a LIDAR Digital Elevation Model (DEM). The proposed algorithm is able to better sharpen the tracing of hydrographic networks compare to the raster DEM methods and to extract drains that were not reported in the known network database. The resulting network better fits the LIDAR network than other raster methods or hydrographic data sources, enhancing surface flow directions and headwater locations. Moreover, it refines the knowledge of hydrographic networks on small catchments by extracting the whole information contained in the TIN structure. Indeed, more than river courses, hydrogeomorphological features are calculated, directly usable in some hydrological models and allowing the study of the relations between local morphology and surface hydrology.