||Characterizing the response of a catchment to rainfall, in terms of the production of runoff vs the interception, transpiration and evaporation of water, is the first important step in understanding water resource availability in a catchment. This is particularly important in small semi-arid catchments, where a few intense rainfall events may generate much of the season's runoff. The ephemeral Zhulube catchment (30 km(2)) in the northern Limpopo basin was instrumented and modelled in order to elucidate the dominant hydrological processes. Discharge events were disconnected, with short recession curves, probably caused by the shallow soils in the Tshazi sub-catchment, which dry out rapidly, and the presence of a dambo in the Gobalidanke sub-catchment. Two different flow event types were observed, with the larger floods showing longer recessions being associated with higher (antecedent) precipitation. The differences could be related to: (a) intensity of rainfall, or (b) different soil conditions. Interception is an important process in the water balance of the catchment, accounting for an estimated 32% of rainfall in the 2007/08 season, but as much as 56% in the drier 2006/07 season. An extended version of the HBV model was developed (designated HBVx), introducing an interception storage and with all routines run in semi-distributed mode. After extensive manual calibration, the HBVx simulation satisfactorily showed the disconnected nature of the flows. The generally low Nash-Sutcliffe coefficients can be explained by the model failing to simulate the two different observed flow types differently. The importance of incorporating interception into rainfall-runoff is demonstrated by the substantial improvement in objective function values obtained. This exceeds the gains made by changing from lumped to semi-distributed mode, supported by 1 000 000 Monte Carlo simulations. There was also an important improvement in the daily volume error. The best simulation, supported by field observations in the Gobalidanke sub-catchment, suggested that discharge was driven mainly by flow from saturation overland flow. Hortonian overland flow, as interpreted from field observations in the Tshazi sub-catchment, was not simulated so well. A limitation of the model is its inability to address temporal variability in soil characteristics and more complex runoff generation processes. The model suggests episodic groundwater recharge with annual recharge of 100 mm year(-1), which is similar to that reported by other studies in Zimbabwe.