We estimate the influence of classmates' ability characteristics on student achievement in exogenously formed university student groups. The study uses administrative data on undergraduate students at a large selective university in Russia. The presence of high-ability classmates has a significant positive effect on individual grades in key economics and mathematics courses as well as on overall academic performance. While a simple linear-in-means model reveals moderate peer effects, non-linear specifications give strong evidence that students at the top of the ability distribution derive the greatest benefit from high-ability classmates. Less able students are not affected by peers and have no significant influence on peers' outcomes.