Discussion of interaction in second language performance assessment have generally been loosely psychological in orientation, forming part of attempts to model the nature of communicative ability within the individual. But in investigating the validity of performance assessments involving interactions between individuals (for example between candidate and interlocutor in speaking assessments), the intrinsically social nature of performance needs to be recognized. What would be the consequences for language testing research if it adopted a social perspective on the nature of interaction? The paper explores the necessity for such a reorientation, and suggests areas that would feature in a consequent research agenda.