This study investigates the roles of collaborative output, the modality of output, and word engagement in vocabulary learning and retention by Chinese-speaking undergraduate EFL learners. The two treatment groups reconstructed a passage that they had read in one of two ways: (1) dyadic oral interaction while producing a written report (Written Output); (2) dyadic oral interaction followed by an oral report (Oral Output). A control group completed a reading comprehension task (Reading) based on the same passage. Four posttests revealed that Oral Output led to significantly better productive and receptive lexical learning than Reading all the way to the last posttest. Written Output led to significantly better productive and receptive lexical learning than Reading on posttest 2, but not on posttests 3 and 4. However, the difference in lexical learning between the Written and Oral Output conditions did not achieve significance. Interaction analysis found that the Oral and Written Output groups differed in the types of word processing they favoured as well as in the frequency of their word engagement. The article discusses the reasons why collaborative output facilitates lexical learning; considers the association between the Output performers' word engagement and lexical retention; and suggests what might have contributed to the better success of the Oral Output group in their lexical retention.