Peer feedback is widely applied to support peer learning and accumulating studies pointed out that feedback features directly impact its learning benefits. However, existing peer feedback studies provide limited insights into group-level peer feedback activities in authentic classrooms. This study conducted group-level peer feedback activity in social studies classrooms of a Singapore secondary school. Fourteen groups of students (N = 61, Female = 61) participated in group-level peer feedback during the computer-supported collaborative argumentation activities. Students' collaborative argumentation and peer feedback were collected. Paired sample t-test was conducted to compare each group's argumentation performance before and after peer feedback activity. Qualitative content analysis was implemented to identify the cognitive and affective features of peer feedback given and received by more-improvement groups and less-improvement groups. A comparison of the feature networks between two student groups revealed the effective practices of peer feedback. The results demonstrated the key role of the specific solution when student groups gave and received peer feedback apart from problem identification and general suggestions. Besides, providing peer feedback at the overall argumentation level was found to be more beneficial than a word or evidence level. When receiving feedback, the use of hedge was found to bring more group improvement than mitigation language. These findings highlight the important features of peer feedback in group-level peer feedback activities, providing insights for the design and instruction of group-level peer feedback activities in authentic classrooms.