Governments around the world use external audits to monitor whether public officials extract rents through shirking on the job or outright embezzlement of public funds. Available evidence on whether such top-down monitoring works is both scant and mixed. While the threat of an audit reduced corruption in local road construction projects in Indonesia (Olken 2007), it had no measurable impact on health worker performance in Ghana (Dizon-Ross et al. 2017). Other topdown monitoring approaches focusing on technological solutions to combat absenteeism in health service delivery have shown similarly mixed results (Banerjee et al. 2008, Dhaliwal and Hanna 2017). Interpretation of these mixed findings is complicated in part because each study evaluates a different intervention in a different context. For example, the government audits in Indonesia might have been effective because prior experience showed that these audits indeed uncovered corruption, while the audits in Ghana were carried out by a nongovernmental organization without established track record. And even identical interventions might have produced different results due to differences in the normative and institutional setting. Yet it is crucial for policymakers to know under what circumstances top-down monitoring can be effective. This paper reports results from a randomized policy experiment that we designed jointly with the Brazilian federal government audit agency (Controladoria-Geral da União, CGU) in order to test whether – holding the context constant – higher audit risk deters rent extraction in three key areas of local government activity: procurement, health service delivery and cash transfer targeting. Annual audit risk increased from the same baseline level of about 5 percent to about 25 percent across the entire range of local government activities. It follows by design that impact heterogeneity across activities in this setting cannot be driven by differences in the intervention itself or by differences in the broader socioeconomic or institutional context. We argue that one should nevertheless expect substantial heterogeneity in response to the same increase in audit risk depending on the parameters of the rent-extraction problem faced by public and private agents involved in these activities. Intuitively, an increase in audit risk alone may fail to substantively reduce rent-taking when sanctions and the probability that they are applied are very low, as is the case in public service delivery in many developing countries (Chaudhury et al. 2006).