ABSTRACT<>A naturalist approach to rationality seems to demand descriptive models. Yet the notion of rationality itself distinguishes between what ought to be and what ought not to be, and so is normative. In this paper, I develop some work on the modeli ng of the role of emotion in rational thought to suggest how the normative character of rationality can be integrated into a naturalistic account. Principles of rationality divide into (strongly or weakly) compulsory and (weakly or strongly) optional. Even in the most compelling cases, however, no rational argument ever compels one to accept a conclusion (since there are alternatives, such as the rejection of a premise, which equally conform to strict logic). Which of the possible alteratives you adopt is irreducibly up to you. In the "most compelling cases", it must be grounded, at some level, on some "wired-in" natural capacity. Such a capacity is analogous to what Pylyshyn calls, in connection with artificial intelligence models, "functional architecture". Functional architecture implements normative rules by virtue of (descriptive) natural laws. Given the distinction between compulsory and descriptive models, however, we clearly can't maintain that all principles of rationality are innate. I hypothesize that in the optional cases the role of functional architecture is often played by emotion, which results in temporary rigidity in the range of admissible relevant considerations. The origin or emotion is in turn a complex mix of natural capacities and social learning; the latter is what gives it its appearance of normativity.