We examine college students' interaction within classrooms and estimate peer effects on their academic performance. We exploit a unique seating rule at a university in South Korea, known as the fixed-seat system. We propose a novel identification strategy based on students' repeated interaction. Our findings show that a student's performance in a class is significantly influenced by his or her next-seat neighbor's ability. The effect is heterogeneous, varying by student and class characteristics. Also quantile regressions reveal that peer effects are significant among below-average students and among those at the top end.