As it was logistically impossible to monitor every event on each plot, estimates of total monsoon soil loss might be made using the regression relationships as predictive tools, with a complete set of monsoon and pre-monsoon rainfall events. However, the highly skewed distribution of rainfall magnitudes and intensities, combined with non-linear runoff and soil loss relationships, and continuous changes in vegetation cover through the cropping cycles, meant that the multiple regression relationships (even with 50% sampling of events of all magnitudes) were not sufficiently well defined, for the highest-magnitude storms, to rely on this technique of estimating soil losses. A longer-term dataset would be required to define, more fully, the mathematical relationships to enable them to be used with a high degree of confidence. Thus, a less sensitive but more robust magnitude and intensity categorization approach was employed to estimate total soil losses from the terraces. Using the total 530 monitored events for 1992 and 1993, a best estimate figure for soil loss was derived for combined classes of rainfall magnitude (40 mm, 40–20 mm, 20– 10 mm, 10–2 mm rain), mean intensity (0.15 mm min–1, 0.15 mm min1) and seasonal vegetation cover (early maize cycle; maize harvest/early millet cycle; and all other times when weed cover 50%) for each terrace.