Many decisions refer to actions that have a causal impact on other events. Such actions allow for mere learning of expected values, but also for causal learning about the structure of the decision context. Whereas most theories of decision making neglect causal knowledge, causal learning theories emphasize the importance of causal beliefs and assume that people represent decision problems in terms of their causal structure. In three studies we investigated the representations people acquire when repeatedly making decisions to maximize a certain payoff. Our results show that (i) initial causal hypotheses guide the interpretation of decision feedback, (ii) consequences of interventions are used to revise existing causal beliefs, (iii) decision makers use the experienced feedback to induce a causal model of the choice situation, which (iv) enables them to adapt their choices to changes of the decision problem.