Behind every theory lies a metatheory, a set of assumptions about the nature of the objects under study, forming a framework within which the theory must operate, and hence largely determining the major focuses of that theory. Here I want to re-examine some of the assumptions that bear most directly on the nature of the object, language, that the theory of generative grammar is meant to explain. Obviously, prior assumptions about the nature of any object of study are unavoidable and not in themselves undesirable. It is impossible to address any topic unless one has at least preliminary notions of what one is dealing with. However, such assumptions can easily form an opaque box, forcing examination of the topic from a particular perspective, and prohibiting examination from any other perspective. The assumptions I shall examine here include the competence-performance distinction, the distinction between E-language and I-language, and the notion that language is (almost) perfect.