The terms ‘downward’ and ‘upward’ (synonymous with ‘top-down’ and ‘bottom-up’ respectively) are sometimes used when describing methods for developing hydrological models. A downward approach is used here to develop a lumped catchment-scale model for subsurface stormflow at the 0·94 km2 Slapton Wood catchment. During the development, as few assumptions as possible are made about the behaviour of subsurface stormflow at the catchment scale, and no assumptions are made about its behaviour at smaller scales. (In an upward approach, in contrast, the modelling would be based on assumptions about, and data for, the behaviour at smaller scales, such as the hillslope, plot, and point scales.) The model has a single store with a relatively simple relationship between discharge and storage, based on equations describing hysteretic patterns seen in a graph of discharge against storage. Double-peaked hydrographs have been observed at the catchment outlet. Rainfall on the channel and infiltration-excess and saturation-excess runoff give a rapid response, and shallow subsurface stormflow gives a delayed response. Hydrographs are successfully simulated for the large delayed responses observed in 1971–1980 and 1989–1991, then a lumped model for the rapid response is coupled to the lumped hysteretic model and some double-peaked hydrographs simulated. A physical interpretation is developed for the lumped hysteretic model, making use of information on patterns of perched saturation observed in 1982 on a hillslope at the Slapton Wood catchment. Downward and upward approaches are complementary, and the most robust way to develop and improve lumped catchment models is to iterate between downward and upward steps. Possible next steps are described. Copyright © 2006 John Wiley & Sons, Ltd.