Two very different theoretical accounts of task-based language use and learning are critiqued and their relevance for language pedagogy discussed. One account, which will be referred to as the psycholinguistic perspective, draws on a computational model of second language (L2) acquisition (Lantolf, 1996). According to this perspective, tasks are viewed as devices that provide learners with the data they need for learning; the design of a task is seen as potentially determining the kind of language use and opportunities for learning that arise. Three different psycholinguistic models are discussed: Long’s Interaction Hypothesis, Skehan’s ‘cognitive approach’ and Yule’s framework of communicative efficiency. The second theoretical account of tasks is that provided by socio-cultural theory. This is premised on the claim that participants co-construct the ‘activity’ they engage in when performing a task, in accordance with their own socio-history and locally determined goals, and that, therefore, it is difficult to make reliable predictions regarding the kinds of language use and opportunities for learning that will arise. Socio-cultural theory emphasizes the dialogic processes (such as ‘scaffolding’) that arise in a task performance and how these shape language use and learning. Both theoretical approaches afford insights that are of value to task-based language pedagogy. The psycholinguistic approach provides information that is of importance for planning task-based teaching and learning. The socio-cultural approach illuminates the kinds of improvisation that teachers and learners need to engage in during task-based activity to promote communicative efficiency and L2 acquisition.