The article examines "task" as research construct as predominantly conceived in terms of task-as-workplan in the task-based learning/second language acquisition literature. It is suggested that "task" has weak construct validity and ontology in an overwhelmingly quantitative paradigm because the construct has a "split personality." Conceptualization is based on the task-as-workplan, but data are gathered from the task-in-process. The article adopts a conversation analysis perspective and demonstrates that the two can be very different. It is proposed that a secure basis for "task" as research construct and for the quantification of discoursal data is attainable only by switching the conceptual and methodological focus to task-in-process.