In this paper the authors describe one aspect of their evolving methodology for providing data for term banks. This methodology is informed by current work in corpus linguistics, and term identifying techniques, which exploit the statistical contrast between the distribution of words in special-language texts and general-language texts. The authors’ future work is shaped by two important considerations. They intend to explore how term banks can reflect the true ‘meaning potential’ of terms, as well as represent the conceptual structure of the domain. The second point concerns the software needs of translators. A future situation is envisaged in which the translator has access to corpora of specialist texts in certain domains and to term banks from the same domains within a unified and integrated software system.