'Normal stress'—a notion frequently encountered in phonology and, especially, syntax—has never been adequately defined. Linguists have apparently made a tacit assumption that the stress in citations elicited from an informant is the same as the stress used by a speaker making a minimum of special assumptions; but this is shown to be false. It is argued that ‘normal stress’, a notion inherited from structuralist linguistics, was required by assumptions inconsistent with those of the generative framework; and that this notion, even if it can be defined so as to be consistent with generative assumptions, is not a particularly useful one.