A study examined global- and local-allocational preselection mechanisms for turn-taking in the language classroom, using two corpuses of classroom discourse illustrating both accuracy- and fluency-based modes of language teaching. It was found that teacher-directed accuracy work yielded Initiation-Response-Feedback patterns governed by local-allocational projection mechanisms for turn-taking, while the fluency-based work was characterized by global-allocational preselection mechanisms for turn-taking, which allow learners to produce multiple-utterance responses. The two corpuses of data display typical discourse features that are related to restricting or facilitating learner initiative, single-utterance versus multiple-utterance learner response, and form versus content feedback. It is shown that these typical discourse features may be used as evidence to confirm or invalidate teacher claims about their modes of language teaching. It is suggested that teachers who are aware of theses features may generate and test specific predictions about their interactions with learners. A set of if-then predictions is generated for each corpus. Data excerpts are appended. Contains 27 references. (Author/MSE)