With computer corpora firmly established as research tools in linguistics, their application for language teaching purposes is also increasingly advocated to the extent that corpus-based language teaching has even been praised as the new revolution in language teaching (cf. Sinclair, 2004b). This article takes a more critical view and examines some of the potential as well as the limitations of computer corpora in the language classroom, providing practical examples from the British Component of the International Corpus of English. It is argued that only a balanced view, which takes into account both the strengths and weaknesses of computer corpora for language teaching, can ensure their successful integration into the language classroom. The discussion first focuses on the limitations of corpus data, which are identified as ‘externalized’, as opposed to ‘internalized’ language, lacking contextual, as opposed to co-textual, information. On the other hand, computer corpora provide access to information not easily available from other sources, viz. information on frequency of occurrence in various text types, and co-occurrence patterns, e.g. collocation, colligation, semantic prosody). This information, however, also has to be seen in the light of the more general questions of how representative corpus results are and to what extent they are generalisable. The article concludes with a discussion of pedagogical implications of the use of computer corpora, especially with regard to their application as tools for exploratory/discovery learning and as means for promoting learner autonomy.