Rainfall-runoff data collected from bare plots (20-216 m2) at 1-min intervals were used to compare the performance of the Green-Ampt infiltration model and a spatially variable infiltration model (SVIM). The two models have the same number of parameters. For 60 natural storm events from six sites in Australia and South-East Asian countries, the average Nash-Sutcliffe model efficiency was 0.77 for the Green-Ampt model and 0.83 for the SVIM. At all sites, the SVIM consistently outperformed the Green-Ampt model when compared to runoff data at a range of time intervals and storm events, including events of very long duration. A larger hydrologic lag is needed for the Green-Ampt model to fit the measured hydrographs in comparison to the SVIM, suggesting that the Green-Ampt model tends to underestimate the infiltration rate when rainfall intensity is high. Measured rainfall and runoff rates show a positive relationship between rainfall intensity and infiltration rate. Considerable spatial variability in the infiltration capacity at the plot scale is implied by this positive relationship. This spatial variability clearly needs to be accommodated in infiltration models, and the SVIM represents a simple formulation of the infiltration rate as a function of rainfall intensity to address this spatial variability. SVIM parameters can be related to the Green-Ampt parameters, and they could therefore be estimated directly using soil properties.