Stated survey measures of risk preferences are increasingly being used in the literature, and they have been compared to revealed risk aversion primarily by means of experiments such as lottery choice tasks. In this paper, we investigate educational choice, which involves the comparison of risky future income paths and therefore depends on risk and time preferences. In contrast to experimental settings, educational choice is one of the most important economic decisions taken by individuals, and we observe actual choices in representative panel data. We estimate a structural microeconometric model to jointly reveal risk and time preferences based on educational choices, allowing for unobserved heterogeneity in the Arrow-Pratt risk aversion parameter. The probabilities of membership in the latent classes of persons with higher or lower risk aversion are modelled as functions of stated risk preferences elicited in the survey using standard questions. Two types are identified: A small group with high risk aversion and a large group with low risk aversion. The results indicate that persons who state that they are generally less willing to take risks in the survey tend to belong to the latent class with higher revealed risk aversion, which indicates consistency of stated and revealed risk preferences. The relevance of the distinction between the two types for educational choice is demonstrated by their distinct reactions to a simulated tax policy scenario.