Smallholder farmers’ land allocation decisions in marginal areas of developing countries typically involve a substantial element of risk, especially when they concern input intensive cash crops. Hence, apart from farmers’ resource endowment, their individual level of risk aversion is a potentially important determinant of such decisions. However, in microeconometric models a measure of individuals’ risk preferences is usually lacking. We address this shortcoming by testing the explanatory power of a wide range of risk preference measures based on hypothetical and non-hypothetical elicitation methods in a model explaining land allocation to commercial hybrid maize production in a fragile upland area of Vietnam. Based on data collected in a random sample of 300 households, we find that the poorest farmers are particularly specialized in commercial maize production, but they are highly dependent on relatively disadvantageous input supply and marketing arrangements offered by maize traders, making this specialization particularly risky. Our study confirms the relevance of decision-makers’ risk preferences in addition to their asset endowment in the land allocation decision. The inclusion of risk preference measures as explanatory variables is found to not cause any significant endogeneity bias. However, only risk preference measures that are based on hypothetical maize related scenarios have explanatory power. We conclude that (1) risk preferences are to a certain extent decision domain specific and (2) hypothetical scenarios that are closely related to farmers’ real-life decisions may produce more reliable results than unfamiliar, non-agricultural scenarios or lottery-based methods, which may be difficult to grasp for respondents with limited formal education.