This paper questions the assumption that corpora are authentic, with particular reference to their application in language pedagogy. The author argues that, because of the form the corpus takes, authentic source texts forfeit a crucial criterion for authenticity, namely context, in the transition from source to electronic data. Other authentic attributes, e.g. communicative intent and socio-cultural purpose likewise become obscured. The problems of attempting to recontextualize corpus discourse and of equating the concept of `authentic¿ with `unique¿ are broached. These dilemmas are resolved by exploiting the pedagogical context itself, by looking not to the authenticity of the source text but to its authentication by the learner, which arises out of the involvement of the learner with the material, via the task. The second part of this paper looks to means of effecting this¿most notably, the methodology known as data-driven learning (DDL) and suggests samples of DDL tasks which accord with this concept of authenticity. © 2004 Oxford University Press.