Modularity is a concept central to cognitive science, and Fodor's analysis of cognitive modularity in his book The Modularity Of Mind has been widely influential - but also widely misunderstood. It is often claimed that the possession of some or other system-property is a necessary condition for that system to be modular in Fodor's sense, but Fodor made it clear that he was not proposing a definition of modularity, nor proposing any necessary conditions for the applicability of the term. He was simply suggesting a number of system properties that are typical of modular systems. I argue that it is nevertheless possible to derive a useful definition of modularity from the kinds of arguments put forward by Fodor: A cognitive system is modular when and only when it is domain-specific. Given any such proposed module, the other features of modularity discussed by Fodor should be dealt with as empirical issues: for each feature (innateness, for example), it is an empirical question whether or not the proposed module has that feature.