To combat money laundering, banks raise and review alerts on transactions that exceed confidential thresholds. This paper presents a data-driven approach to detect smurfing, i.e., money launderers seeking to evade detection by breaking up large transactions into amounts under the secret thresholds. The approach utilizes the notion of a counterfactual distribution and relies on two assumptions: (i) smurfing is unfeasible for the very largest financial transactions and (ii) money launderers have incentives to make smurfed transactions close to the thresholds. Simulations suggest that the approach can detect smurfing when as little as 0.1-0.5\% of all bank transactions are subject to smurfing. An application to real data from a systemically important Danish bank finds no evidence of smurfing and, thus, no evidence of leaked confidential thresholds. An implementation of our approach will be available online, providing a free and easy-to-use tool for banks.