A key issue in modelling large agent systems is *what* to model. This has an effect on how we represent models and execute them. In previous talks I have described specifications of models which map closely to the structure of deployed agent systems - they allow a reasonably direct mapping between the structure of the model and the structure of the system we are modelling. There are, however, many forms of more abstract model which may be more appropriate in certain circumstances. I shall give an initial catalogue of those which I know about and attempt to compare them informally in terms of the different insights they might give into model (and hence system) behaviour.