Structuralism sought to introduce various kinds of autonomy into the study of language, including the autonomy of that study itself. The basis for this was the insistence on categorial autonomy, whereby categories are identified language-internally (whether in a particular language or in language). In relation to phonology, categorial autonomy is tempered by grounding: the categories correlate (at least prototypically) with substance, phonetic properties. This is manifested in the idea of ‘natural classes’ in generative phonology. Usually, however, in more modern grammars, despite some dissent, no such grounding (in meaning) has been attributed to syntax. This attitude culminates in the thesis of the ‘autonomy of syntax’ which was put forward in transformational-generative grammar. In what follows here it is argued that the consequences of this are very unfortunate. Distribution alone is insufficient to determine the identity of categories; what is relevant is the distribution of the prototypical members of the category, where prototypicality is notionally defined. Prototypical nouns, for instance, denote concrete, discrete, stable entities. Syntax, as well as phonology, is grounded. Groundedness ensures that only the prototypical behaviour of semantically prototypical members of a category determines its basic syntax; and this syntax reflects the semantic properties. Groundedness filters out potential syntactic analyses that are incompatible with this. For instance, given the diverse semantic characters of prototypical nouns and verbs, groundedness predicts that the X-bar theory of syntactic structure, which attributes parallel projections to lexical categories, is false. Consideration of the syntax of nouns and verbs confirms that this is indeed the case. The attribution to syntax of categorial autonomy without grounding should be abandoned.