This paper addresses a lack of evidence on the impact of performance pay in the public sector by evaluating a pilot scheme of incentives in a major government agency. The incentive scheme was based on teams and covered quantity and quality targets, measured with varying degrees of precision. We use data from the agency's performance management system and personnel records plus matched labour market data. We focus on three main issues: whether performance pay matters for public service worker productivity, what the team basis of the scheme implies, and the impact of the differential measurement precision. We show that the use of performance pay had no impact at the mean, but that there was significant heterogeneity of response. This heterogeneity was patterned as one would expect from a free rider versus peer monitoring perspective. We found that the incentive scheme had a substantial positive effect in small teams, and a negative response in large teams. We found little impact of the scheme on quality measures, which we interpret as due to the differential measurement technology. We show that the scheme in small teams had non-trivial effects on output, and our estimates suggest that the use of incentive pay is much more cost effective than a general pay rise.