In this paper, we argue that language teaching which aims to foster speaking skills and natural spoken interaction should be based upon the grammar of spoken language, and not on grammars which mainly reflect written norms Using evidence from a mim-corpus of conversational English, we look at how four grammatical features which occur with noticeable frequency in the corpus are dealt with in currently popular pedagogical grammars Our investigation shows that treatment of the selected features varies from adequate to patchy to complete absence from the grammars surveyed We conclude that research in discourse analysis does offer some helpful insights into the usage of these features, but that, in the absence of easy access to discourse analysis work and given the mixed treatment in grammar books, teachers and learners will often be thrown back on their own resources However, we argue that even very small amounts of real spoken data can yield significant evidence which can be used imaginatively within inductive and language awareness approaches in and out of the classroom to increase awareness and knowledge of the grammar of conversation