Is belief ascription automatic? Some findings suggest that it is, others that it is not. Can infants ascribe false beliefs? Some measures indicate that they can, others that they cannot. Reflection on these and other puzzling patterns of findings suggests that we might usefully ask, What it is to ascribe beliefs? More generally, What it is to be a mindreader? Mindreading involves representing mental states. And just as representing physical states requires having some model of the physical, so equally representing mental states requires having some model of the mental. Now the history of science makes it plain that there are multiple models of the physical; some models are relatively easy to apply but are limited in accuracy or scope, others are more accurate but also harder to use. My talk starts from the observation that there are also multiple models of the mental. To say that someone represents beliefs or other mental states leaves open the question of which model of the mental she is using. By outlining how to construct a minimal theory of mind, I shall show that there are simple but limited models of the mental. Perhaps, then, mindreading sometimes involves simple but limited models of the mind. This might contribute to resolving puzzles about automaticity and development. But how could hypotheses about which model a mindreader is using be tested? Different models have different signature limits, and these limits may make it possible to identify the operation of a given model across contexts and across types of subject. These limits generate predictions, making it possible to test the conjecture that mindreading involves not only multiple systems but also multiple models of the mind.