Poverty in Kenya has been on the increase over the last decade. It is estimated that 56% of the total population live in absolute poverty. Most of the poor reside in rural areas where agriculture is the main livelihood activity. Majority of Kenya's farmers are smallholders. They account for 75% of the total agricultural output and about 70% of marketed agricultural produce. The fate of smallholder agriculture in this country is therefore central to poverty reduction. A collapse in output and incomes from smallholder agriculture is likely to have damaging welfare effects and retard pro-poor economic growth. Smallholders often operate in a risky environment, which affects the level and variability of household resource endowments and income. The importance of risk analysis in smallholder production systems arises from the fact that strategies to help farmers raise their productivity and income require an understanding of how risk affects their production decisions. Using cross-sectional input-output data from 240 randomly selected households from Vihiga and Kilifi districts covering the 2003/4 agricultural year, and time series data on yields, this study investigates how risk affects farmers' production choices in the crop-livestock systems. Further, it explores the possibility of improving production and income on these smallholder farms. Linear programming and MOTAD are used in analysis. Results indicate there is potential to improve production and income on these smallholder farms through a change in the enterprise mix. The farm plans are sensitive to risk, with trade-offs occurring between higher risk and returns.