The heterogeneity in farmer crop portfolios observed empirically in a given location is puzzling. Even in the presence of higher-return alternatives, a large number of households focus on low-yield food crops, and portfolio choices vary among apparently similar households. This article aims at understanding better crop portfolio choices in Burkina Faso. A theoretical framework emphasizes the relationship between risk and crop portfolio, and shows that household characteristics may lead to different diversification choices. The empirical study area is composed of nine villages in the region of Nouna (Western Burkina Faso), characterized by poor infrastructures, high levels of risk and poverty, and market failures. Quantitative and qualitative information was collected among farmers and traders during a five-month fieldwork in 2008-2009, providing data for a mixed-methods analysis. The quantitative dataset includes 229 household-year observations, which correspond to 933 plot-year observations. Several measures of diversification are used: the number of crops cultivated, the Theil (or entropy) index, and the Berry and McVey (BM) index. Probit and multinomial logit models are employed to identify the determinants of diversification, as well as the choice of particular crops. Results are robust and show that overall, better-off households have more diversified crop portfolios. While poorer households grow mostly basic food crops (millet and sorghum), richer households include higher-yield food and cash crops (cotton, maize and rice) in their portfolio, a situation leading to crop-based poverty traps for poorer households. Among the main cash crops, cotton is cultivated by the richest households, while sesame is introduced as a diversification crop by relatively poorer households, benefitting from favorable market conditions. The findings contradict the general perception that the poorest households diversify more in order to cope with agricultural shocks. In addition to risk, several factors can explain household diversification: high levels of risk and uncertainty even for better-off households; economies of scale; high transaction costs; and input and equipment constraints. Overall, the results suggest two main types of policy interventions. First, social safety nets such as cash transfers and index insurance can be employed to help households cope with risk and extend their portfolios toward more profitable crops. Second, infrastructure development and trading platforms can be used to improve market conditions. The findings also call for more research related to crop portfolio choices, trader behavior and market conditions in rural Sub-Saharan Africa.