The present study takes as an empirical point of departure the nature of interaction in second language speaking tests. We examine the relationship between ratings of students’ performance in an oral proficiency test and the social practice of conducting ‘test talk’. Using conversation analysis, our focal point is how students in peerdriven test interactions manage trouble related to the task-at-hand. Given that students were assessed not only on their linguistic skills, but also on their interactional ability and treatment of topics assigned, our emphasis on task management stems from a hypothesis that orientation to the test task is intimately connected to overall test outcome. We demonstrate that different types of task-related trouble (TRT) reveal diverse understandings of the test task and that ‘doing-being a successful task manager’is connected to a moderate orientation to the task and test format. Students displaying such task management strategies were also assessed as highly proficient, whereas other task management strategies identified in our study correlated with low scores and grades. However, the relationship between subskill ratings and task management was not always clear-cut. We argue that the diverging understandings of the test task that learners display become part of how they are assessed and that certain task management strategies are rated less favorably than others. Our study holds promise for the fine-tuning of oral proficiency subskill ratings and raises questions as to the framing of test tasks, since this appears to have implications for student performance and evaluation.