Abstract This paper presents the development and application of a distributed rainfall–runoff model for extreme flood estimation, and its use to investigate potential changes in runoff processes, including changes to the ‘rating curve’ due to effects of over-bank flows, during the transition from ‘normal’ floods to ‘extreme’ floods. The model has two components: a hillslope runoff generation model based on a configuration of soil moisture stores in parallel and series, and a distributed flood routing model based on non-linear storage–discharge relationships for individual river reaches that includes the effects of floodplain geometries and roughnesses. The hillslope water balance model contains a number of parameters, which are measured or derived a priori from climate, soil and vegetation data or streamflow recession analyses. For reliable estimation of extreme discharges that may extend beyond recorded data, the parameters of the flood routing model are estimated from hydraulic properties, topographic data and vegetation cover of compound channels (main channel and floodplains). This includes the effects of the interactions between the main channel and floodplain sections, which tend to cause a change to the rating curve. The model is applied to the Collie River Basin, 2545 km 2, in Western Australia and used to estimate the probable maximum flood (PMF) from probable maximum precipitation estimates for this region. When moving from normal floods to the PMFs, application of the model demonstrates that the runoff generation process changes with a substantial increase of saturation excess overland flow through the expansion of saturated areas, and the dominant runoff process in the stream channel changes from in-bank to over-bank flows. The effects of floodplain inundation and floodplain vegetation can significantly reduce the magnitude of the estimated PMFs. This study has highlighted the need for the estimation of a number of critical parameters (e.g. cross-sectional geometry, floodplain vegetation, soil depths) through concerted field measurements or surveys, and targeted laboratory experiments.