The ﬁeld of cognitive neuropsychology employs the patterns of performance observed in brain-damaged patients to constrain our models of normal cognitive function. This methodology was historically based upon simple “box and arrow” models, with particular cognitive deﬁcits being taken as indicative of the selective breakdown of corresponding “boxes” or “arrows”. In principle, within this framework, one should be able to piece together a complete model of mental structure by studying patients with complementary patterns of deﬁcit (e.g. Caramazza, 1986; Shallice, 1988). The concept of double dissociation has been of particular importance for this enterprise, with its presence being taken to imply modularity across a whole range of systems. I shall review the technical details and provide speciﬁc examples later, but the basic inference is that if one patient can perform task 1 but not task 2, and a second patient can perform task 2 but not task 1, then a natural way to explain this is in terms of separate modules for the two tasks.