We propose a localization approach in urban environments based on correspondences between street photographs and a 2D GIS. Assumption of vertical facades and use of vanishing points theory allows extracting a partial geometry from the images : angles and width ratios between facades. This information is then transformed into GIS queries, which result in a layer of every possible pose for the camera. Wrong poses are then discarded based on consistency criteria, using road layer information or building layer information that has not been used before. Eventually, combination of possible poses coming from several images of the same site result in a unique pose, the true one. We show results of our experiments based on real data of Nantes city, France