With the development of information technology, co-viewing of live video streaming (LVS) has become a popular online learning method. However, existing studies have found inconsistent results regarding the effects of co-viewing, which could be due to the impact of learner-learner interactions. The present study tested the effects of co-viewing LVS on learning in elementary students, and whether learner-learner interaction moderated students' attention allocation, learning performance (i.e., retention and transfer), learning efficiency, and metacognition. The study used a one-way between-subjects design, with 86 participants assigned randomly to one of three groups: learning alone group, merely co-viewing group, or co-viewing with interaction group. Kruskal-Wallis H tests showed that students in the co-viewing with interaction group allocated more attention to their co-viewer and less to the LVS. However, ANOVA results indicated that they had the best learning performance and metacognition, and demonstrated the highest learning efficiency. Meanwhile, those co-viewing without interaction did not show significantly positive effects compared to those learning alone. The results of informal interviews were largely consistent with the above findings. The findings of the present study suggest the benefits of co-viewing with interaction, providing practical implications for the social context of learning from LVS for elementary students in particular.