Flood simulations demand mathematical models, which are rigorously calibrated and validated against benchmarking datasets. For this purpose, experiments are conducted in a river-network-floodplain set-up. Hypothetical stepped hydrographs are passed through the channel-network, and fluvial flooding situations are created. Flood depths are recorded at various locations and evolving flood extents are extracted by image processing. TELEMAC 2D is tested against the observed data. The most accurate scheme for flood prediction is identified through sensitivity analysis. Inclusion of the turbulence model is found to improve the accuracy in predicting dynamic flood extents. The model seems to slightly overpredict inundation extents during the rising limb of the hydrographs and underpredict during the falling limb. In addition, certain aspects of a flood such as river–floodplain interaction and junction hydraulics cannot be reproduced with high precision by the 2D model. The experimental datasets can be a valuable resource to mathematical modellers and are freely downloadable.