Most work on the evolution of cognition assumes that cognition is a representational form of information processing that arises in circumstances in which there is selection for behavioral flexibility. It is common in empirical research to define cognition operationally in terms of behavior that cannot be accounted for by associative learning. However there is no consensus on the nature of the representational information processing that marks the emergence of cognition, nor is there any clarity on how it might enable enhanced behavioral flexibility. This a crucial gap that needs to be filled: research on animal cognition and its evolution is diverse and productive, but it sorely lacks a theoretical framework that poses core issues in a systematic way. Empirical research is fragmented across many fields that use a diverse range of concepts for characterizing cognition. Two areas of research are especially promising: work on the hippocampus and its role in spatial cognition (Jacobs 2003) and declarative memory (Eichenbaum 2000, 2004), and work on the neural mechanisms subserving goal-directed behavior (Yin and Knowlton 2006). This research casts light on the fundamental mechanisms of representational flexibility and their neural substrate. However its nature and significance are not widely appreciated in animal cognition research generally, and there is currently no theoretical analysis that develops appropriately generalized concepts that could serve as a synthetic framework for the diverse areas of research in the evolution of cognition. The main objective of this project is to develop just such a theoretical analysis. It will elaborate a concept of model-based representation and show that it is an appropriate focus for understanding the emergence of representationally mediated behavior control. Model-based representation can be regarded as a generalization of Tolman’s ‘cognitive map’ concept. The analysis will outline the nature of model-based representation, show some of the kinds of flexibility that it provides, and characterize computational mechanisms likely to be involved in the formation of representational models. Some steps will be taken towards showing that this concept can serve as a framework for integrating a diverse range of findings on animal cognition; in addition to the research mentioned above, it will address research on animal conceptual abilities, and work on behavioral innovation and problem solving.