Much evidence has accumulated in favor of such a dual view of reasoning (Evans, 2003, in press; for arguments against, see Osman, 2004). There is however some vagueness in the way the two systems are characterized. Instead of a principled distinction, we are presented with a bundle of contrasting features - slow/fast, automatic/controlled, explicit/implicit, associationist/rule based, modular/central - that, depending on the specific dual process theory, are attributed more or less exclusively to one of the two systems. As Evans states in a recent review, “it would then be helpful to have some clear basis for this distinction”; he also suggests that “we might be better off talking about type 1 and type 2 processes” rather than systems (Evans, in press). We share the intuitions that drove the development of dual system theories. Our goal here is to propose in the same spirit a principled distinction between two types of inferences: ‘intuitive inference’ and ‘reflective inference’ (or reasoning proper). We ground this distinction in a massively modular view of the human mind where metarepresentational modules play an important role in explaining the peculiarities of human psychological evolution. We defend the hypothesis that the main function of reflective inference is to produce and evaluate arguments occurring in interpersonal communication (rather than to help individual ratiocination). This function, we claim, helps explain important aspects of reasoning. We review some of the existing evidence and argue that it gives support to this approach.