The selection of an efficient physical schema is an NPcomplete problem. In this paper, we show that crucial parts of physical database design can be smoothly modelled as a Dempster-Shafer application. We exploit the properties of the Dempster-Shafer theory to model explicitly a rich set of heuristics —used for the selection of an efficient physical schema— into knowledge rules. These rules may be loaded into a knowledge base, which, in turn, can be embedded in database design tools.