To map water levels in case of large floods, it is proposed to use GIS to manage the great amount of information extracted from aerial photographe. Our method is developed in 3 stages: 1) segmentation of the flood plain into sectors with acceptable mean water depth, based on geographical limite and size criteria; 2) determination of minimum and maximum depth for each sector, based on the emergence or not of natural objects (vegetation, dikes ...); 3) amelioration of the estimation of the water depths, using all observed hydraulic connections between sectors (cracked dikes, flows ...). This amelioration is done solving a constraint system linking the différent estimated water levels. Application concerne the Hérault River (France) for the November 1994 flood, with aerial photographs taken 6 hours after the maximum of flood. Results give good determination about water levels, the lateral dynamic behaviour of the flood (1.5 m of difference in levels between centre and edges of plain), and the impact of dikes. These results are obtained without introducing either general hydraulic equations nor local head loss coefficients. The great quantity of fuzzy information, compensates for its low quality.