Previous research has documented that collaborative dialogue promotes L2 learning in both face-to-face (F2F) and synchronous computer-mediated communication (SCMC) modalities. However, relatively little research has explored modality effects on collaborative dialogue. Thus, motivated by sociocultual theory, this study examines how F2F compares with SCMC regarding the generation of collaborative dialogue specifically in terms of its frequency and nature. Thirty-two Chinese English as a foreign language (EFL) students participated in this study and completed two types of collaborative tasks (i.e. dictogloss and jigsaw) in dyads in both modalities. The analyses of learners’ exchanges focused on language-related episodes (LREs), the instantiation of collaborative dialogue. The identified LREs were categorized based on their focus, outcome and type. A follow-up questionnaire was conducted to elicit students’ perspectives. The results revealed that LREs were more frequent in SCMC than in F2F. Furthermore, the analyses of the nature of LREs indicated some cross-modality differences: whereas SCMC LREs had the features of orthographical, correct and self-correction outcomes, F2F LREs were characterized by incorrect and request for assistance outcomes. Pedagogical implications and suggestions for the future research were also discussed.