This study develops a methodology for finding metaphors in corpora. The procedure is based on the wish that, without a prior list of metaphors, the computer would provide a number of possible metaphor candidates. The methodology works by selecting an initial pool of word types in the corpus, finding shared collocates between pairs of those words and then computing a semantic distance measure for those word pairs which have a requisite number of mutual collocates. Cases which satisfy these criteria were then concordanced and interpreted. This methodology was applied to a corpus of MA dissertations in Applied Linguistics, completed in Brazil. The paper highlights the importance of the use of metaphors by novice Applied Linguistic researchers.