We exploit the institutional setting of the EU preliminary ruling mechanism to separate out team and agency motives for inter-court cooperation. Nonhierarchical referral regimes, we argue, reflect a pure team model. However, agency problems may occasionally arise when the work of the referring courts fail to meet the referral court's quality standards. To address these agency costs, the referral court may use formal dismissals. We demonstrate that, depending on the strength of the referring court's team motive, prior beliefs and reputational and learning costs, the negative feedback conveyed by the referral court's dismissal can either produce (1) a chilling effect inhibiting subsequent cooperation, or (2) a learning effect. Using data on preliminary references in the EU legal system, we design separate tests for the two hypotheses and implement these in multilevel models. We specifically model learning as a function of lagged exposure to formal dismissal using a Delayed Lagged Model (DLM). We reject the chilling hypothesis but find support for the learning hypothesis and the team model.