LAST decade witnessed the burgeoning interest among lexicographers and linguists in using computer-based authentic corpora and concordances in codifying the social behaviour of language. The effect of corpora on linguistic descriptions is such that it is at least practically impossible to disregard “evidence” (Sinclair, 1990, 2004) anymore in favour of “intuition” (Chomsky). That is, the role of evidence in both language description and teaching is now elevated to the level where the generalizations the learners make are “corpus-driven” rather than “corpus-based” (Biber and Conrad, 1990). In order for the corpus driven methodologies to be successful in enabling the learners draw valid generalizations the corpus they consult needs to unbiased, large enough and representative. COCA (Corpus of Contemporary American English) is one such large (450 million words) representative corpus of current English of multiple varieties. COCA being a general corpus of contemporary American English provides the users a range of possibilities: a provision for comparable corpora such as BNC, COHA, TIME, a provision for a separate academic corpus of 110 million words within the 450 million word database, a provision for a spoken corpus etc. The present paper attempts to closely examine the academic component of COCA for academic vocabulary (Cox head, 2000) in the light of the “phraseology” theories proposed by Sinclair (1990). Academic corpora will be studied for patterns of word behaviour across disciplines: education, science, law…; and finally based on the analysis pedagogic implications will be suggested. Language Corpora and Learning Opportunities: An Analysis of the Academic Corpus of Contemporary American English (COCA-A) LAST two decades have witnessed the burgeoning interest among lexicographers and linguists in using computer-based ‘real language data’ in understanding and codifying the language in use. The effect of corpora on linguistic descriptions is such that any attempt to disregard “evidence” (Sinclair, 1991)in favour of intuitive judgements leads to “hopelessly unreliable” (Sampson, 2001) generalizations about language use. Regarding any evidence as sacrosanct too however reduces the role of real language as doubly hopelessly unreliable. The composition of real data to deduce ‘valid’ generalizations about language in use thus forms the major concern of the paper. The present paper closely examines the pedagogic potential of the academic component of COCA in learning‘academic vocabulary’(Coxhead, 2000)(Hyland & Tse, 2007). Academic corpora is studied for patterns of word behaviour; and finally based on the analysis pedagogic implication sare suggested. © Research India Publications.