Many psychologists, philosophers, and computer scientist have written about mental models, but have remained vague about the nature of such models. Do they consist of propositions, concepts, rules, images, or some other kind of mental representation? This paper will argue that a unified account can be achieved by understanding mental models as representations consisting of patterns of activation in populations of neurons. The fertility of this account will be illustrated by showing its applicability to causal reasoning and the generation of novel concepts in scientific discovery and technological innovation. I will also discuss the implications of this view of mental models for evaluating claims that cognition is embodied.