As in-field data acquired during a flood are often limited to point information and are spatially sparse, available 2- or 3-dimensional mathematical hydraulic models dramatically suffer from a lack of suitable calibration and validation datasets. Remote sensing appears to be a very powerful alternative acquisition mode that could address some of these difficulties. After a brief review of literature on water level extraction using remote sensing, a methodology is proposed that aims to extract water levels from aerial photographs taken during flooding. It consists of estimating the water levels in hydraulic compartments over the whole inundated plain at the moment the photographs were taken. The idea developed is that the processing of remote sensing images taken during flooding can go beyond a simple and descriptive extraction of the inundated area if it incorporates ancillary data and if it is guided by hydraulic functioning. The methodology is presented in three steps: (1) the extraction of local water-level estimates; (2) the extraction of a spatial structure composed of compartments and hydraulic circulations; (3) the combination of these data to improve water- level estimates. The methodology, illustrated here on a Herault site but also tested on a portion of the Moselle and Aisne Rivers, has shown its efficacy in providing quantitative and distributed data on water levels. These have proven sufficiently precise (mean +/- 15 cm on the three tests carried out) to be used and assimilated in numerical hydraulic modelling.