This study draws on complete friendship network data on two first-year biological sciences cohorts at a selective university in the United States to investigate how and to what extent allocating students to curricular groups and grading their performance in class shape (1) processes of friend selection at the dyadic level and (2) friendship clustering at the network level. Through a set of stochastic actor-oriented models, results show that students tend to befriend peers from the same curricular group versus a different one (i.e., curricular group homophily) and befriend higher-performing peers (i.e., performance-based status). Follow-up analyses reveal that friendship clustering by curricular group placement is largely due to course co-enrollment (i.e., proximity), whereas academic-performance-based clustering is primarily the result of students aligning their own performance to match the average performance of their friends (i.e., influence). I discuss implications of these findings for helping to promote learning in higher education.