The portrayal of the actions of fictive characters for purposes of entertainment is a familiar phenomenon. Theories that seek to explain why we are attracted to such fictions and whether we learn from them have produced no consensus and no adequate overall account. In this paper, we present the hypothesis that entertainment relies on cognitive adaptations for pretend play. As a simplified model system, we draw on our field study of children's chase play, which is characterized by an elementary form (...) of pretense. The children pretend, at first without consciously representing their pretense, to be chased by predators. The details of this behavior, widespread among mammals, indicate that the biological function of the game may be to train predator-evasion strategies. Chase play, we suggest, evolved in early mammals because it enabled cheap and plentiful resources to be used to train strategies for events that are rare, dangerous, and expensive. More generally, we argue that pretense is used to access spaces of possible actions in order to locate and practice new strategies. It relies on the creation of a simulated scenario and requires sophisticated source monitoring. The simulation is experienced as intrinsically rewarding; boredom is a design feature to motivate the construction of a more appropriate pedagogical situation, while the thrill of play signals optimal learning conditions. The conscious narrative elaboration of chase games involves an elementary form of role play, where we propose a virtual agent is created that tracks and acts on the memories required for coherent action within the simulation. These complex if familiar design features, we suggest, provide a minimalist functional and adaptationist account of the central features of entertainment: that it is fun, that it involves us imaginatively and emotionally, and that it has a tacit pedagogical effect. The model provides a principled and testable account of fiction-based entertainment grounded in evolutionary and cognitive processes. (shrink)