Using experiments on explicit vs. implicit teaching methods as a starting-point, this paper raises some questions on corpora and second-language acquisition: To what extent are controlled experiments arranged to test the effectiveness of the use of corpora in language teaching? To what extent can the use of corpora be grounded in theories of language acquisition and of language in general? Attention is drawn to the connection between ideas of learning as hypothesis testing as expressed by proponents of error analysis and the more recent ideas connected with corpora and the student as a researcher. Reference is made to the usage-based model, which may account both for the nature of language and language acquisition. The usage-based model and the relevance of corpora deserve to be recognised in work on second-language acquisition. It is claimed that corpora have many uses of relevance for language teaching and language learning. Special attention is drawn to three types of corpora: textbook corpora, learner corpora, and multilingual corpora. There is a need, however, for more systematic studies on the effectiveness of corpora in language teaching. Their use is vindicated to the extent that it agrees with what we know about language and language acquisition, and can be shown to be an effective learning tool.