The social capital theory reveals the importance of peer relationships on students’ learning. However, it is unclear how students select their collaborators under the influence of their previous collaborations and backgrounds. This study explores to what extent students’ free selection choices for collaborators among their peers are based on previous collaboration in formally structured groups (i.e., learning communities (LCs)) and based on different students’ background characteristics. A parallel program was studied where students studied in one of four LCs for two years and after that, they have to find their own group members within or across LCs to finish their bachelor thesis in the third year. In total, 1152 students’ selections of their peers were analyzed. This paper presents the percentages of students choosing group members within or across LCs. It also considered the influence of students’ backgrounds, like sex, nationality, and academic performances on their peerchoices by logistic regression analysis. More than half of the students chose group members within their own LC, regardless of which LC they were in. Although the majority of the students chose collaborators within their own LC, still around 40% of students were willing to collaborate with others from different LCs with whom they had never collaborated before in the formal curriculum. Students’ backgrounds (i.e., sex, and academic performance) were also associated with their decisions. A high frequency of collaboration within formally structured groups enhances the students’ preference of group members from the same groups, but also informal peer relationships are crucial in students’ choices for collaboration. Students’ sex and academic performance influence their free choice of group members while nationality does not. Students with different academic levels have a higher chance to become group members when they collaborated before in formally structured groups than those students who had not had such a collaboration experience.