Attracting and retaining high-quality teachers has a large social benefit, but it is challenging for schools to identify good teachers ex-ante. This paper uses teachers’ contract choices and a randomized controlled trial of performance pay with 7,000 teachers in 243 private schools in Pakistan to study whether performance pay affects the composition of teachers. Consistent with adverse selection models, we find that performance pay induces positive sorting: both among teachers with higher latent ability and among those with a more elastic effort response to incentives. Teachers also have better information about these dimensions of type than their principals. Using two additional treatments, we show effects are more pronounced among teachers with better information about their quality and teachers with lower switching costs. Accounting for these sorting effects, the total effect of performance pay on test scores is twice as large as the direct effect on the existing stock of teachers, suggesting that analyses that ignore sorting effects may substantially understate the effects of performance pay.