Estimation of direct runoff volume and peak runoff rates for small rural catchments in Kenya, East Africa, poses a problem in view of the limited gauged data available. The problem can be circumvented by adopting a modelling approach. This, however, does require some base data for model calibration and validation. A highly reliable data base for two small catchments, Sambret (7.02 km2) and Lagan (5.44 km2), for nearly 22 years (1958–1980) is used in the present study. These catchments lie in the headwaters of the Lake Victoria drainage basin in Kenya and were intensively instrumented by the former East African Agricultural and Forestry Research Organisation (EAAFRO). In the Sambret catchment, 41 rainfall-runoff events were available, out of which 25 were used for model calibration while the other 16 were used for validation. To further verify the applicability of the models to different catchments, 18 events from the neighbouring Lagan catchment were also used. The SCS curve number procedure proposed by the United States Soil Conservation Service was used to estimate the direct runoff volumes. The curve numbers were derived from storm rainfall-runoff relationships and the results obtained were satisfactory. Peak runoff rates were derived by both Nash and kinematic wave models for the estimated runoff volumes. The Nash model gave acceptable results in both Sambret and Lagan. The kinematic wave model performed well in Sambret only; in Lagan, it under-predicted the peak runoff rates.