Two experiments, one conducted with American college students and one with Israeli pilots and their instructors, explored the predictive power of reputation-based assessments versus the stated "name of the game" (Wall Street Game vs. Community Game) in determining players' responses in an N-move Prisoner's Dilemma. The results of these studies showed that the relevant labeling manipulations exerted far greater impact on the players' choice to cooperate versus defect--both in the first round and overall--than anticipated by the individuals who had predicted their behavior. Reputation-based prediction, by contrast, failed to discriminate cooperators from defectors. A supplementary questionnaire study showed the generality of the relevant short-coming in naïve psychology. The implications of these findings, and the potential contribution of the present methodology to the classic pedagogical strategy of the demonstration experiment, are discussed.