Task-based language assessment (TBLA) grows from the observation that mastering the grammar and lexicon of a language is not sufficient for using a language to achieve ends in social situations. In TBLA, language use is observed in settings that are more realistic and complex than in discrete skills assessment and that typically require the integration of topical, social, and pragmatic knowledge along with the knowledge of the formal elements of language. But designing an assessment is not accomplished simply by determining the settings in which performance will be observed. TBLA raises questions of just how to design complex tasks, evaluate students' performances, and draw valid conclusions from them. This paper examines these challenges from the perspective of"evidence-centered assessment design." The main building blocks are student, evidence, and task models, with tasks to be administered in accordance with an assembly model. These models are described, and how they are linked and assembled frame an assessment argument is demonstrated. Points are illustrated with examples from TBLA. (Contains 2 figures and 48 references.) (Author/SLD)