A novel field experiment shows that learning activities in pairs with a greater spread in abilities lead to better individual work performance, relative to those in pairs with similar abilities. The positive effect of the former is not limited to their performance in peer learning material, but it also spills over to their performance in other areas. The underlying improvement comes from the stronger increased performance of those whose achievements were weak prior to peer learning. This implies that exogenously determining learning partners with different abilities helps improve productivity through knowledge sharing and potential peer effects.