This paper compares the predictions made by two competing general models of phrase structure for cross-language word order patterns. The first general model is categorial grammar, in which constituents contract ‘function-argument’ relations. The second is exemplified by theories such as the X-bar theory in which constituents are either heads of phrase or (various types of) modifiers of these, and hence in which the ‘modifier-head’ relation holds. It is the general differences between these two types of relations which will concern us, for they result in partially different structural groupings being assigned to the same surface constituents, and in very different node labellings. In particular ‘they embody different claims about the kinds of syntagmatic relations that hold between constituents. Functionargument relations are primarily semantically motivated, and the syntactic rules of phrasal construction are assumed to mirror these isomorphically. Modifier-head relations are typically syntactically motivated, within the context of autonomous theories of syntax. The set of word order data to be used is taken from Hawkins (1983). The question we shall address is: how well does one or the other theory succeed in describing and explaining these universal word order data? There are, of course, numerous exemplars of each of these models. Compare, for example, the categorial grammars of Ajdukiewicz (1935), Montague (1974), Vennemann (1976, 198 l), Vennemann and Harlow (1977), Keenan (1979), Keenan and Faltz (1978), Flynn (1982). Modifier-head