Standard economic models assume that the weight given to information from different sources depends exclusively on its diagnosticity. In this paper we study whether the same piece of information is weighted more heavily simply because it arose from direct experience rather than from observation. We investigate this possibility by conducting repeated game experiments in which groups of players are randomly rematched on every round and receive feedback about the actions and outcomes of all players. We find that participants' actions are influenced more strongly by the behavior of players they directly interact with than by those they only observe.