Economists study data on choices that people make and from this deduce people's preferences and expectations. This identification process, as well as the predictions based on it, becomes flawed when multiple sets of preferences and expectations are consistent with the same data. One way forward would be to measure directly people's expectations on future states of the world. This paper discusses the theoretical merits and practical constraints of doing so. Data on risk perceptions seem particularly relevant for understanding savings and investment behavior in the developing world, where risk is pervasive and often posited to have significant costs. Although the main aim of the paper is thinking through the 'whys' and the 'hows' of measuring risk perception, some interesting, but still unrepresentative findings from the field are presented. For example, we uncover a puzzle in that respondents seem to be engaged in farming activities that are stochastically dominated by other farming activities. Furthermore we find evidence that suggests that for high-profile activities, like owning a shop, the heuristic of availability makes people without experience in the activity bias their distributions to the right compared to people with experience.