We provide results of a randomized trial comparing incentives for improved delivery of public services in India, targeting child nutrition through the work of salaried caregivers in Chandigarh, India. A winner-take-all prize paid to the best performer yielded less improvement than dividing the same award among workers in proportion to their share of measured gains. In our population of about 2,000 children served by 85 workers, using proportional rewards led to weight-for-age malnutrition rates that were 4.3 percentage points lower at 3 months (when rewards were paid) and 5.9 points lower at 6 months (after the contest had ended), with mean weight-for-age z scores that were .071 higher at 3 months, and .095 higher at 6 months. Proportional bonuses led to larger and more sustained gains because of better performance by lower-ranked workers, whose efforts were not rewarded by a winner-take-all prize. Results are consistent with previous laboratory trials and athletic events, demonstrating the value of proportional rewards to improve service delivery for child nutrition and other development outcomes. Acknowledgement : This project was funded by the Bill & Melinda Gates Foundation (OPP1107973), with additional support from Amherst College and the Bharat Prakarsh Foundation. We are especially grateful to the mothers, children and Anganwadis who participated in our trial, to the Chandigarh Social Welfare Department and Child Development Bureau for their support, to our survey staff especially Alka Yadav, Paulin Priscilla and Sam Alpert, and for very helpful comments from Jere Behrman, Sonia Bhalotra, Karthik Muralidharan and Alessandro Tarozzi and other participants at the Conference on Child Development at the UPenn Center for Advanced Studies on India (CASI), 15-16 September 2017.