For a three-year time period beginning in 2001, North Carolina awarded an annual bonus of $1800 to certified math, science and special education teachers working in public secondary schools with either high-poverty rates or low test scores. Using longitudinal data on teachers, we estimate hazard models that identify the impact of this differential pay by comparing turnover patterns before and after the program's implementation, across eligible and ineligible categories of teachers, and across eligible and barely-ineligible schools. Results suggest that this bonus payment was sufficient to reduce mean turnover rates of the targeted teachers by 17%. Experienced teachers exhibited the strongest response to the program. Finally, the effect of the program may have been at least partly undermined by the state's failure to fully educate teachers regarding the eligibility criteria. Our estimates most likely underpredict the potential outcome of a program of permanent salary differentials operating under complete information. © 2007 Elsevier B.V. All rights reserved.