When students can self-select their group members, a common assumption is that students prefer to select friends from similar cultural backgrounds. However, when teachers randomize students in groups from different cultural backgrounds, students are "forced" to work together. The prime goal of this study is to understand the impact of two group selection methods on how students from diverse cultural backgrounds build learning and work relations, using an innovative quantitative method of Social Network Analysis in a pre-post test manner. In a quasi-experimental study of 2 x 69 students, in one condition the students were randomly allocated to groups by staff and in the other, students were allowed to self-select their group members. The results indicate that students in the self-selected condition primarily selected their friends from a similar cultural background. The learning networks after 14 weeks were primarily predicted by the group allocation and initial friendships. However, students in the random condition developed equally strong internal group relations but more "knowledge spillovers" outside their group, indicating that the random condition led to positive effects beyond the group.