We use a global games approach to model alternative implementations of an antitrust leniency program as applied to multiproduct colluders. We derive several policy design lessons; e.g., we show that it is possible that linking leniency across products increases the likelihood of conviction in the first product investigated but reduces it in subsequent products. Thus, firms may have an incentive to form sacrificial cartels and apply for leniency in less valuable products to reduce convictions in more valuable products. Cartel profiling can mitigate this undesirable effect, but also reduces the probability of conviction in the first product investigated. (JEL D43, D86, K21, L12, L41)