We use information on firms’ trading networks from VAT return data to design a randomized tax compliance intervention in Uganda. In treated pairs, either the seller, the buyer, or both receive letters listing discrepancies detected in past tax returns. The amendment rate is 22 percentage points higher in the treatment group, compared to 1.8% in the control group. We find spillover effects within treated firm pairs and in transactions with their untreated trading partners. Overall, there is a small increase in VAT liability for the amended returns. The intervention also leads to fewer discrepancies in subsequent tax declarations.