This paper is about decision making based on real-world descriptions of a domain. There are many domains where different people have described various parts of the world at different levels of abstraction (using more general or less general terms) and at different levels of detail (where objects may or may not be described in terms of their parts) and where models are also described at different levels of abstraction and detail. However, to make decisions we need to be able to reason about what models match particular instances. This paper describes the issues involved in such matching. This a the basis of a number of fielded systems that do qualitative-probabilistic matching of models and instances for real-world data.