This paper reports from an on-going study on interaction in paired speaking tests.The main data consists of video recorded speaking tests in the final national test in Swedish for adult immigrants (B1/B1+ on the CEFR-scale). The participants have varying ethnic, linguistic and educational backgrounds. Oral proficiency is generally considered difficult to assess in a valid and reliable way. The rationale for using the paired format (a candidate-candidate discussion) is that it elicits a wider range of speech functions and results in more complex talk compared to traditional oral proficiency interviews (Brooks 2009). Meanwhile, there are also problems linked to the format; the interlocutor effect is hard to control and giving individual scores in a co-constructed interaction is a challenge (McNamara 1997). The theoretical framework for the national test in Swedish for immigrants draws on different models of communicative competence (eg. Celce-Murcia 2007). Interactional competence is a key feature in the theoretical framework underpinning the test. Candidates are tested through interaction (by a candidate candidate discussion) and on interaction (by criteria on interaction set out in the syllabus). How is interactional competence manifested in a test setting? Can different proficiency levels of interactional competence be described and assessed? Can different contextual recourses (Linell 2009), like prior schooling and background knowledge, have an impact on the understanding of the situation and on the interaction? Preliminary findings indicate that the candidates use different interactional resources and strategies during the test. Different proficiency levels of interactional competence cannot be easily described and assessed.