A burgeoning recent literature emphasizes "livelihood" diversification among smallholder populations (Chambers and Conway 1992, Davies 1993, Ellis 1998, Bryceson 1999, Ellis 2000, Little et al. 2001). While definitions vary within this literature, the concept of livelihoods revolves around the opportunity set afforded an individual or household by their asset endowment and their chosen allocation of those assets across various activities to generate a stream of benefits, most commonly measured as income. This holistic perspective has the potential to enhance our understanding of the strategies that farm households pursue to ensure food and income security given the natural and economic environment in which they operate. Diversification patterns reflect individuals' voluntary exchange of assets and their allocation of assets across various activities so as to achieve an optimal balance between expected returns and risk exposure conditional on the constraints they face (e.g., due to missing or incomplete markets for credit, labor, or land). Because it offers a glimpse as to what people presently consider their most attractive options, given the incentives and constraints they face, the study of diversification behavior offers important insights as to what policy or project interventions might effectively improve either the poor's asset holdings or their access to higher return or lower risk uses of the assets they already possess. Since diversification is not an end unto itself, it is essential to connect observed livelihood strategies back to resulting income distributions and poverty. Not all diversification into off-farm or non-farm income earning activities offers the same benefits and not all households have equal access to the more lucrative diversification options. Yet the livelihoods literature offers little documentation or explanation of important differences between observed diversification strategies. This paper addresses that gap by offering a comparative analysis using data from three different countries, Cote d'Ivoire, Kenya and Rwanda. Like Dercon and Krishnan (1996) and Omamo (1999), we emphasize that interhousehold heterogeneity in constraints and incentives must factor prominently in any sensible explanation of observed diversification behaviors. Indeed, section 4 demonstrates that at a very fundamental level - the choice of basic livelihood strategy - households would prefer locally available livelihood strategies other than those they choose, were they not constrained from doing so. A simple appeal to the principle of revealed preference thus suggests that heterogeneous constraints and incentives play a fundamental role in determining livelihood diversification patterns manifest in income diversification data. The plan for the remainder of this paper is as follows. The next section presents the basic conceptual foundation from which we operate. Section 3 then introduces the data sets and definitions employed in the analysis. Section 4 presents findings relating to the observed variation in income sources across the income distribution, to distinct livelihood strategies pursued by rural African households, to the determinants of strategy choice, and to the effects of alternative livelihood strategies on income dynamics. These findings point especially to significant rural markets failures - especially with respect to finance and land - that force poorer subpopulations to select strategies offering demonstrably lower returns while wealthier subpopulations are able to enjoy higher return strategies to which entry is at least partly impeded by fixed costs and lower marginal costs of participation. Section 5 concludes.