The standard analysis of treatment effects only considers the gross benefit of treatment and does not consider the cost as perceived by the agents or the surplus arising from participation in programs (the private subjective evaluation of the program). This paper extends the analysis of Heckman and Vytlacil (1999, 2005) to identify parameters measuring the costs and net gains of arising from participation in a program within the context of a generalized Roy model. The analysis does not require that the analyst observe the subjective cost of treatment. Instead, we use information derived from agent choices about participating in a program to infer the cost of treatment as perceived by the agent. We apply our methodology to the analysis of college choice and nd that variability in net gains from attending college is mainly due to variability in the cost associated with schooling.