Pay-for-performance is a popular public education reform, but there is little evidence about the characteristics of a well-designed incentive pay plan for teachers. Some of the literature suggests that effective incentive plans must offer relatively large awards to induce behavioral changes. On the other hand, the experimental economics literature suggests that plans with only a handful of awardees can be less effective at changing behavior than plans that offer an array of possible awards. Still other research suggests that group-based incentives are the most effective strategy when teamwork and cooperation are integral to the production process--as is arguably the case in education. This study takes advantage of a pilot pay-for-performance program in Texas to explore incentive design not only from the perspective of the employer--by examining changes in teacher productivity and retention--but also from the perspective of the employee--by examining the preferences revealed by the incentives teachers design for themselves. We find that when given the opportunity, teachers design relatively weak, group-oriented incentive pay plans. In turn, those relatively weak incentives do not appear to be associated with any significant changes in teacher productivity, although they are correlated with teacher turnover, which, in the long run, could theoretically improve student outcomes.