Despite the preponderance of theoretical and empirical evidence that suggests the use of pair/group work to promote second language learning, it is still unclear who can best form high performance groups. Should students be allowed to choose their working partners, or should teachers themselves assign students to pairs? This study set out to compare the nature of student-selected and teacher-assigned pairs while they were engaged in collaborative writing. All learner talk was audio recorded, transcribed and analysed for the quantity, type and resolution of language related episodes (LREs) as well as the patterns of dyadic interaction. Furthermore, the study examined the texts produced using both quantitative and qualitative measures. Our findings suggest that the teacher-assigned pairs generated significantly more LREs than the student-selected pairs, while there was no significant difference in the patterns of interaction between the two pairing methods. Meanwhile, the qualitative analysis of learner talk revealed a considerable amount of off-task behavior among the members of student-selected pairs. Moreover, as far as the outcome of pair work (collaborative writing) was concerned, the teacher-assigned pairs noticeably outperformed the student-selected pairs on measures of fluency and accuracy. Also, they produced significantly better texts in terms of organization, grammar and vocabulary.