In a randomized trial conducted with primary school students in China, we find that pairing high and low achieving classmates as benchmates and offering them group incentives for learning improved low achiever test scores by approximately 0.265 standard deviations without harming the high achievers. Offering only low achievers incentives for learning in a separate trial had no effect. Pure peer effects at the benchmate level are not sufficiently powerful to explain the differences between these two results. We interpret our evidence as suggesting that group incentives can increase the effectiveness of peer effects. © 2014 Elsevier B.V.