Scientific theories are often accepted on grounds other than empirical adequacy, by employing non-empirical factors such as the simplicity, elegance or scope of the theory. In fact, often scientists commit to theories that have not yet received empirical confirmation, basing their confidence on its non-empirical virtues. How can we justify the non-empirical factors employed in theory choice? Can we defend the idea, often put forward by scientific realists, that the virtues of a theory are indicators of its truth? In this paper I revisit arguments defending the idea that the virtues of a theory have epistemic import. Following Bas van Fraassen’s account of theory choice, I argue that the non-empirical factors employed in theory choice should best be seen as pragmatic devices for choice, reflecting not the ontology of the world, but rather our own practices and aims.