Medical decision-support systems in which uncertainty plays an essential role are increasingly based on the formalism of probabilistic networks. Although this formalism is very powerful, the construction of actual networks is not straightfor- ward, and requires the availability of clearly structured medical domain models as a starting point. In this paper it is argued that medical pathophysiological knowledge constitutes a good start for the development of such models, even though pathophysiological knowledge is semantically different from probabilistic knowledge. Two models concerning anaemia, which are part of a broad system covering the domain of anaemia, are discussed to illustrate the general approach.