TOPMODEL, a semi-distributed, topographically based hydrological model, was applied to simulate continuously the runoff hydrograph of a medium-sized (379 km2), humid tropical catchment. The objectives were to relate hydrological responses to runoff generation mechanisms operating in the catchment and to estimate the uncertainty associated with runoff prediction. Field observations indicated that water tables were not parallel to the surface topography, particularly at the start of the wet season. A reference topographic index REF was therefore introduced into the TOPMODEL structure to increase the weighting of local storage deficits in upland areas. The model adaptation had the effect of deepening water tables with distance from the river channel. The generalized likelihood uncertainty estimation (GLUE) framework was used to assess the performance of the model with randomly selected parameter sets, and to set simulation confidence limits. The model simulated well the fast subsurface and overland flow events superimposed on the seasonal rise and fall of the baseflow. The top ranked parameter sets achieved modelling efficiencies of 0·943 and 0·849 in 1994 and 1995 respectively. The GLUE analysis showed that the exponential decay parameter m, controlling the baseflow and the local storage deficit, was the most sensitive parameter. There was increased uncertainty in the simulations of storm events during the early and late phases of the season, which was due to a combination of: errors in detecting the rainfall depths for convectional rainfall events; the treatment of rainfall as a catchment areal value; and, the strong seasonality in runoff response in the humid tropics.