Against a background of recent progress in developmental neuroscience, some of which has been taken as challenging to the modularity hypothesis of Fodor (1983), this article contrasts two competing conceptions of modularity: sui generis modularity, according to which modules are treated as independent neurocognitive entities that owe nothing to one another, and descent-with-modification modularity, according to which current cognitive modules are understood to be shaped by evolutionary changes from ancestral cognitive modules. I argue that sui generis modularity is incompatible with a range of data, from the co-occurrence of deficits to the patterns of activation in neuroimaging studies, but that same range of data is compatible with descent-with-modification modularity. Furthermore, I argue that the latter conception of modularity may have important implications for the practice and conception of fields such as developmental disorders and linguistics.