The recent flash flood that occurred in the South of France showed that the flooding of roads is a major problem for two reasons: i) it is a significant source of danger for car passengers ii) emergency services need to have a clear overview of the road conditions to be efficient. Therefore it appears important to determine the points of a road network liable to be submerged during flash floods. The objective of this study is to progress in the forecasting of road flooding. The Gard region has been taken as the application case study to develop and to test the proposed method. The Gard region is frequently affected by severe flash floods and an inventory of the road points flooded during a forty year period, the PICH, has recently been realized in part of the region which can serve as a support for the calibration and the validation of the method. The proposed approach consists in three steps. First, a vulnerability rating method, taking into account the characteristics of the road and of its upstream watershed has been calibrated and validated using the PICH information. In a second step a semi-distributed rainfall-runoff model has been developed to be able to evaluate flood discharges produced by the upstream watershed of all the points of the road network exposed to flooding. Finally, a method has been proposed to combine the vulnerability level of a point of the road network and the computed discharges to be able to compute a flooding risk level at each moment of a flood event and at every location of the road network exposed to flooding. The approach has shown promising results on five recent events for which maps of observed road submersions were available.