Storage cell codes for inundation modelling treat the floodplain as a series of discrete basins, with the flow between cells calculated using some analytical flow formulae such as the Manning equation. These codes have many of the advantages of full two-dimensional schemes but without the computational cost. However, with this class of model the propagation speed of the inundation front over the floodplain can be shown to be highly dependent on the model grid scale and insensitive to floodplain friction. To overcome these problems adaptive time step techniques have recently been proposed; however, to date these have only been tested against highly idealised analytical solutions. In this paper the authors compare fixed and adaptive time step storage cell codes for flood inundation modelling against a real world data set for the first time. The data consist of a satellite synthetic aperture radar image of inundation extent at ∼12·5 m spatial resolution taken during a 1-in-50-year flood, which occurred in 1998 on the upper River Severn in the UK; gauged flows at the upstream and downstream ends of an approximately 60 km reach of the river; and a digital elevation model developed from channel survey and an airborne laser altimeter survey of the floodplain topography. Results show that the adaptive model produces a modest improvement in simulation performance over the fixed time step code, yet is much better able to simulate realistically the drying of the floodplain in the upper reaches of the model domain. The adaptive code also demonstrates a more realistic sensitivity to floodplain friction which is likely to account for its markedly different dynamic behaviour and, in this respect, intuitively better performance.