We conduct a survey and find that 47% of respondents state they would sanction free riders in a team production scenario even though the respondent was not personally affected and no direct benefits could be expected to follow an intervention. To understand this phenomenon, we define social reciprocity as the act of demonstrating ones disapproval, at some personal cost, for the violation of a widely-held norm (for example, dont free ride). Social reciprocity differs from reciprocity because social reciprocators punish all norm violators, regardless of group affiliation or whether or not the punisher bears the costs. Social reciprocity also differs from altruism because, while the latter is an outcome-oriented act benefiting someone else, the former is a triggered response not conditioned on future outcomes. To test the robustness of our survey results, we run a public goods experiment that allows players to punish each other. The experiment confirms the existence of social reciprocity and additionally demonstrates that more socially efficient outcomes arise when reciprocity can be expressed socially. Further we find that most subjects who punish do so to discipline transgressors and helping others is largely a positive externality. Finally, to provide some theoretical foundations for social reciprocity, we show that generalized punishment norms survive in one of the two stable equilibria of an evolutionary public goods game with selection drift.