The notion of ‘modularity’ has been given many interpretations in cognitive science, in Fodor (1983) and elsewhere. Which of those notions should be employed, in order that a massively modular account of human cognitive architecture can be defensible? The way to answer this question is to consider the main arguments that have been offered in support of massively modular accounts of the mind, extracting from them a matching account of modularity. These are the arguments from biological design generally, the argument from the organization of learning / animal minds, and the argument from computational tractability. The account of modularity that emerges is weaker than Fodor’s (in particular, modules needn’t be encapsulated), but nevertheless substantive enough to be interesting. Viz.: the mind consists of a great many distinct and dissociable processing systems, all of which are computationally frugal in their operations, and with internal processing that is inaccessible elsewhere.