Economists and neuroscientists often explain game playing by assuming that humans try to predict the opponent's behavior on the basis of her past choices. We try to question this assumption in a Prisoner's Dilemma Game by using a methodology which we call the “subtractive behavioral method”. Our aim is to investigate which task features make participants attend to the opponent's behavior or, on the contrary, make them take into account only their own choices and received payoffs. We find a critical effect of contextual information and we derive some suggestions about the methodology of brain imaging and behavioral game theory experiments.