In sparsely cropped farming systems in semi-arid tropics, rainfall partitioning can be complex due to various interactions between vertical and horizontal water flows, both in the atmosphere and in the soil. Despite this, quantifying the seasonal rainfall partitioning is essential, in order to identify options for increased yields. Results are presented on water flow components, based on field measurements and water balance modelling, for three years (1994–96) in a farmer's field cultivated with pearl millet [ Pennisetum glaucum (L.) Br.] in the Sahel (Niger). Water balance modelling was carried out for three common infiltration categories: runoff producing surfaces, surfaces receiving inflow of runon water from upstream zones, and a reference surface with zero runoff and runon. Runoff was calculated to 25%–30% of annual rainfall (which ranged from 488 to 596 mm), from crust observations, rainfall, soil wetness data, and infiltration estimates. Inflow of runon was estimated from field observations to 8%–18% of annual rainfall. The parameters in the functions for soil surface and canopy resistances were calibrated with field measurements of soil evaporation, stomatal conductance and leaf area. The model estimates of soil water contents, which were validated against neutron probe measurements, showed a reasonable agreement with observed data, with a root mean square error (RMSE) of approximately 0.02 m 3 m −3 for 0–160 cmsoil depth. Estimated productive water flow as plant transpiration was low, amounting to 4%–9% of the available water for the non-fertilised crop and 7%–24% for the fertilised crop. Soil evaporation accounted for 31%–50% of the available water, and showed a low variation for the observed range of leaf area (LAI <1m 2 m −2). Deep percolation was high, amounting to 200–330 mm for the non-crusted surfaces, which exceeded soil evaporation losses, for 1994–95 with relatively high annual rainfall (517–596 mm). Even a year with lower rainfall (488 mm) and a distinct dry spell during flowering (1996), resulted in an estimated deep percolation of 160 mm for the non-fertilised crop. The crop did not benefit from the additional inflow of runon water, which was partitioned between soil water storage and deep percolation. The only exception to this was the fertilised crop in 1996, where runon somewhat compensated for the limited rainfall and the higher water demand as a result of a larger leaf area than the non-fertilised crop.The effects of rainfall erraticness, resulting in episodic droughts, explain why a crop that uses such a small proportion of the available water, in an environment with substantial deep percolation, still suffers from water scarcity. Application of small levels of phosphorus and nitrogen roughly doubled yields, from 380 to 620 kg ha −1, and plant transpiration, from 33 to 78 mm. Evapotranspirational water use efficiency ( WUE ET) was low, 6500–8300 m3 ton-1 grain for non-fertilised crop,which is an effect of the lowon-farmyields and high non-productive water losses. The estimated seasonal rainfall partitioning indicates the possibility of quantifying vertical water flows in on-farm environments in the Sahel, despite the presence of surface overland flow.