The performance of a two-dimensional numerical model of flood hydraulics is tested for a major event in Carlisle, UK, in 2005. This event is associated with a unique data set, with global positioning system surveyed wrack lines and flood extent surveyed 3 weeks after the flood. The simple finite-volume model is used to solve the two-dimensional Saint-Venant equations over an unstructured mesh of 30 000 elements representing channel and floodplain, and allowing detailed hydraulics of flow around bridge piers and other influential features to be represented. The simple finite-volume model is also used to corroborate flows recorded for the event at two gauging stations. Calibration of Manning's n is performed with a two-stage strategy, with channel values determined by calibration of the gauging station models, and floodplain values determined by optimising the fit between model results and observed water levels and flood extent for the 2005 event. The root mean square error for the calibrated model compared with surveyed water levels is of the order of ±0·4 m, the same order of magnitude as the estimated error in the survey data. The study demonstrates the ability of unstructured mesh hydraulic models to represent important hydraulic processes across a range of scales, with potential applications to flood risk management.