Abstract: This paper analyzes peer effects at a large European university across six cohorts of undergraduate economic students. Students are randomly assigned two groups of peers. The university fosters bonds only within one of these groups. These two peer groups, close and distant, form a student’s tutorial group whom they follow classes with throughout the first year. We find the existence of positive peer effects that originate from close peers only. This implies spillovers arise due to social interactions and not due to classroom-level effects. Towards the end of the first year the peer effects disappear. We argue these dynamics arise because students interact less with their randomized peers as time goes on. Using administrative data on tutorial attendance in first year, we find students cluster more with their close peers in the beginning than in the end of the year. Using open tutorial registration in second year, we find that students co-register with only one extra student from their close peer group. Shared gender, ethnicity, and former bonds are stronger predictors of co-registration. Policies that exploit spillovers should focus on social interaction, though long-term benefits are unlikely with endogenous selection of peers.