Changes in the nature of higher education are leading towards increased interest in the assessment of student learning. This study considers an attempt to apply value-added models for the purposes of comparing student learning across institutions, taking care to discuss special considerations inherent to the application of these models to higher education. Using standardised outcome data from Colombia on 64,000 students, this study focuses on the amount of ability sorting that exists and the effect of model specification and choice of outcome on the subsequent results. The level of student ability clustering into colleges was shown to be comparable to both primary/secondary schools and higher education institutions in the United States. As in other studies, models that controlled for contextual effects tended to result in very different rankings of institutions as compared to models without contextual effects. Rankings were approximately as sensitive to the choice of outcome (subject specific vs. generic) as the choice of model, although the subject-specific models did explain more of the overall varation.