The authors test the power of loss aversion to improve teacher performance. During the 2010-11 Chicago school year teachers were randomly asked to participate in a pay-for-performance program with “gain” and “loss” treatments. The “gain” group received traditional financial incentives at the end of the year in the form of bonuses linked to student achievement. Those teachers in the “loss” group were paid a lump-sum in advance and asked to give back the money if their students did not meet performance targets. Teachers in both conditions received the same monetary bonus if they reached the same performance targets.This approach resulted in increases in math test scores for the loss condition by an equivalent of increasing teacher quality by more than one standard deviation. The gain treatment yields smaller and statistically insignificant results. The authors attribute the significant difference between the loss and gain condition to the loss aversion framing.