This paper examines the determinants of the choice of the major when the length of studies is uncertain, by using a framework in which students entering post-secondary education are assumed to anticipate their future earnings. For that purpose, we use French data coming from the 1992 and 1998 Génération surveys collected by the Centre d'Etudes et de Recherches sur l'Emploi et les Qualifications (CEREQ, Marseille). Our econometric approach is based on a semi-structural three-equations model, which is identified thanks to some exclusion restrictions. We exploit in particular exogenous variations in the earnings returns associated with the majors across the business cycle, in order to identify the causal effect of expected earnings on the probability of choosing a given major. Relying on a three-component mixture distribution, we account for correlation between the unobserved individual-specific terms affecting the preferences for the majors, the unobserved individual-specific factors entering the equation determining the length of studies within each major, and that affecting the labor market earnings equation. Following Arcidiacono and Jones (2003), we use the EM algorithm with a sequential maximization step to produce consistent parameter estimates. Simulating for each given major a 10 percent increase in the expected earnings suggests that expected earnings have a statistically significant but quantitatively small impact on the allocation of students across majors.