In this paper we use an economic model to analyse data from a major social experiment, namely PROGRESA in Mexico, and to evaluate its impact on school participation. In the process we also show the usefulness of using experimental data to estimate a structural economic model. The evaluation sample includes data from villages where the program was implemented and where it was not. The allocation was randomised for evaluation purposes. We estimate a structural model of education choices and argue that without such a framework it is impossible to evaluate the effect of the program and, especially, possible changes to its structure. We also argue that the randomized component of the data allows us to identify a more flexible model that is better suited to evaluate the program. We find that the program has a positive effect on the enrollment of children, especially after primary school; this result is well replicated by the parsimonious structural model. We also find that a revenue neutral change in the program that would increase the grant for secondary school children while eliminating for the primary school children would have a substantially larger effect on enrollment of the latter, while having minor effects on the former.