Scientists have been dissecting the neural architecture of the human mind for several centuries. Dissecting its computational architecture has proven more difficult, however. Within the cognitive sciences, for example, there is a debate about the extent to which human reasoning is generated by computational machinery that is domain specific and functionally specialized. While some claim that the same set of cognitive processes accounts for reasoning across all domains (e.g. Rips 1994, Johnson-Laird & Byrne 1991), others argue that reasoning is generated by several different mechanisms, each designed to operate over a different class of content (e.g. Baron-Cohen 1995, Cheng & Holyoak 1985, Cosmides & Tooby 1992, Leslie 1987). Indeed, it has recently been proposed that the human cognitive architecture contains a faculty of social cognition; a suite of integrated mechanisms, each of which is specialized for reasoning and making decisions about a different aspect of the social world. Candidate devices include a theory of mind mechanism, an eye direction detector, social contract algorithms, permission schemas, obligation schemas, precaution rules, threat detection procedures and others (e.g. Baron-Cohen 1995, Cheng & Holyoak 1985, Cosmides 1985, 1989, Cosmides & Tooby 1989, 1992, 1994, Fiddick et al 1995, Fiske 1991, Jackendoff 1992, Leslie 1987, K. Manktelow & D. Over, unpublished paper, 1st Int Conf on Thinking, Plymouth, UK 1988, Manktelow & Over 1990, M.Rutherford, J. Tooby, L. Cosmides, unpublished paper, 8th Annual Meeting Human Behav Evol Society, Northwestern Univ, IL 1996, J. Tooby & L. Cosmides, unpublished paper, 2nd Annual Meeting Human Behav Evol Society, Evanston, IL 1989). To decide among these sometimes competing proposals, psychologists need empirical methods and theoretical standards that let us carve social inference mechanisms at the joints. We will argue that the theoretical standards needed are those of the 'adaptationist programme' developed in evolutionary biology. To show how these standards can be applied in dissecting the computational architecture of the human mind, we will discuss some recent empirical methods and results.