In this paper, we investigate the extent to which knowledge compilation can be used to improve model checking and inference from propositional weighted bases. We first focus on the compilability issue for both problems, deriving mainly non-compilability results in the case preferences are subject to change. Then, we present a general notion of