This paper illustrates the use of large text databases for acquiring new knowledge about the structure of language, summarizing the results of investigations by Justeson and Katz (1991), and elaborating upon issues concerning the textual basis of the research. Our particular problem is to characterize the nature of the relation of ANTONYMY. This relation is a facet of the lexical semantics of all languages We show that corpora are not only convenient and useful for investigating manifestations of this relation, but are in fact crucial to its very definition (at least for English and we suppose for other languages as well). In particular, we use a 25,000,000-word database to establish a new definition for antonymy. This result shows that corpus concerns are critical not only to practical issues in natural language processing, but also to elements of theoretical linguistics. © 1992 Oxford University Press.