Lexicography is currently in the doldrums, due in part to the collapse of traditional business models for funding lexicographical research, which used to be dependent on predicted sales of printed books (i.e. dictionaries). If lexicography is to have a future, it needs to show its usefulness in new domains, notably language teaching and computational linguistics. To do this successfully, it will have to develop radical new approaches to accounting for words and meanings. This paper investigates one such area of potential usefulness, namely the relationship between phraseology and meaning. Generative linguists such as Beth Levin (1993, 2005) have long argued, or rather assumed, that each word has a meaning and that this meaning determines its possible uses. Lexicographers, on the other hand, examine usage in order to find out what a word means. In recent years, examination of usage has been greatly facilitated by the advent of large electronic corpora. The prediction of pre-corpus systemic functional linguists such as Firth (1957) and Sinclair (1966) that word usage is highly patterned have been borne out by innumerable detailed studies of usage. Do such studies support the conclusion that meaning determines possible phraseology? Or do they instead support the conclusion that phraseology determines meaning? The relationship between meaning and usage raises theoretical questions that are of great concern to lexicographers and empirical linguists alike. There is clearly a relationship between meaning and patterns of usage, but the directionality is uncertain. What determines what? The bulk of the paper will be taken up (insofar as time allows) with discussion of some theoretical questions and implications raised by findings of the procedure called corpus pattern analysis (Hanks 2004, 2013). Examples include: • The role of domain in determining possible meaning: e.g., is a lawyer filing a lawsuit doing the same thing as an admin assistant filing papers in a filing cabinet? And what about a journalist filing a story or a pilot filing a flight plan? • The role of subargumental clues in determining (or selecting) meaning. Consider the ambiguity of he took his place; contrast the unambiguous statement She took his place. What about other collocations of take + place? • When is ellipsis possible? If you can say “He fired” and mean “He fired a bullet from a gun”, why can’t you say “The university fired” and mean “The university fired some employees from their jobs”? • What is the relationship between verbs and nouns in making meanings? A preliminary discussion of the relationship between nouns and verbs in making meanings can be found in Hanks (2012). Among the tentative conclusions of the paper are that meaning does not determine usage, but nor does usage determine meaning. Instead, the relationship is symbiotic. A person who engages in linguistic behaviour chooses to make active use of patterns of usage (and sometimes to exploit them in unusual ways), in order to realize some sort of communicative and/or social intention. Central to this enterprise is reliance on (or exploitation of) stereotypes of usage patterns in order to realize stereotypical (and sometimes, original newly created) meanings. It would be useful to compare the stereotype theory of Hilary Putnam (1975) with the prototype theory of Eleanor Rosch (1978). Are they compatible or, perhaps, even identical? To understand in detail how meaning works, it will be necessary to map prototypical meanings onto prototypical patterns of usage. A product of such a mapping would be lexicographical: an inventory of stereotypical patterns of usage, each of which would be associated with a meaning (presuppositions and entailments) – or a translation into a corresponding idiomatic pattern in a foreign language. Such an inventory would serve as a basis for pattern matching, of the kind proposed by Yorick Wilks as long ago as 1973. Wilks’s work offers useful guidance for interpreting texts, in the form of the aphorism “Best match wins”. Wilks’s problem in 1973 was that no inventory of patterns existed for any language. Sadly, this is still the case. But the difference is now we have sufficient evidence, in the form of large corpora, to enable people to build such inventories, for many languages. It is up to lexicographers to build them, and up to rich software houses (or foundations, or research funding agencies) to find them.