Flood mapping proved to be feasible using multitemporal ERS SAR amplitude images. However, it allows to map the flood extent only at given acquisition dates, rarely corresponding to the maximal flood extent. In this paper, we report first results of a study on the use of coherence images, in order to improve the flood detection which can be difficult in some flood conditions. We focused our study on a major flood that occured in 1996 in southern France. In this area was available a large data set including water height measurements, DEM's, different thematic vector layers in a GIS and Tandem INSAR ESR images. Then, we compared the results of flood detection, obtained with using respectively amplitude and coherence images, to flood simulations derived from a simple hydraulic model. We showed that coherence images should detect areas where flood-induced surface changes occured, possibly corresponding to the maximal flood extent within the repeat-pass acquisition window.