Debate continues about what existing evaluations of the impact of Problem Based Learning tell us and about how future evaluations should be designed. This paper argues that establishing descriptive causation should be an important goal of such evaluations. Minimizing threats to internal validity is of key importance to establishing descriptive causation. It is argued that whilst not a panacea randomized experiments offer particular advantages in this respect. This is illustrated through comparison of examples of evaluations of Problem Based Learning that have used different types of study design.