This study investigated the extent to which the use of interactional resources (i.e., turn-taking, topic management, repair and task management) and the co-construction of interactional patterns (i.e., collaborative, parallel, asymmetric and blend) differed in paired speaking tests when EFL test-takers who had the same vs. different proficiency levels were paired together. The study was carried out with 100 EFL learners paired as low-low, high-high and low-high in an oral proficiency exam conducted at a Turkish state university. The results revealed that high-high pairs created a collaborative pattern in which pairs collaboratively extended the topic, supported each other, and had balanced turn-taking opportunities, satisfactory other-initiated topic moves and task-related management. On the contrary, low-low pairs usually created a parallel pattern in which pairs had balanced turn-taking opportunities, but limited topic extension, listener support, and task-related management. Lastly, low-high pairs generated an asymmetric pattern due to the dominance of the high-level test-takers. This study suggests that when pairs are matched with a different proficiency level partner, the communicative and affective advantages of paired speaking tests may be compromised in a way to put both lower and higher proficiency groups at risk due to the asymmetrical interactional pattern co-constructed by these pairs. (C) 2019 Published by Elsevier Ltd.