W hen the Allied forces invaded Nor- mandy during the Second World War, thousands of people were involved in the preparations and in the inva- sion itself; a similar number of Germans were probably involved in defending the occupied territory. War is a prime example of large-scale within-group cooperation between genetically unrelated individuals (Fig. 1). War also illustrates the fact that within-group cooperation often serves the purpose of between-group aggression. Modern states are able to enforce coopera- tion in large groups by means of sophisticated institutions that punish individuals who refuse to meet their duties and reward those who follow their superiors’ commands. The existence of such cooperation-enhancing institutions is very puzzling from an evolu- tionary viewpoint, however, because no other species seems to have succeeded in establishing large-scale cooperation among genetically unrelated strangers 1 . The puzzle behind this cooperation can be summarized as follows. Institutions that enhance within-group cooperation typically benefit all group members. The effect of a single group member on the institution’s success is negligible, but the contribution cost is not negligible for the individual.Why, therefore, should a self-interested individual pay the cost of sustaining cooperative insti- tutions? More generally, why should a self- interested individual contribute anything to a public good that — once it exists — an individual can consume regardless of whether he contributed or not? On page 499 of this issue 2 , Panchanathan and Boyd sub- stantially advance the scope of reputation- based models 3–5 and show that individuals’ concern for their reputation may be a solu- tion to this puzzle. Evolutionary psychologists have sought to answer the puzzle of human collective action for decades. However, progress was limited because of a lack of commitment to mathematically rigorous theorizing. Many researchers erroneously thought that Trivers’s notion of reciprocal altruism 6 , which Axelrod and Hamilton successfully formalized as a tit-for-tat strategy for two- person interactions, provides the solution to the problem. Trivers himself speculated that reciprocal altruism “may favour a multiparty altruistic system in which altruistic acts are dispensed freely among more than two individuals”. However, it is always easier to speculate than to provide a rigorous model, and the speculation is likely to be wrong in this case. In the context of the problem of public- goods provision, a reciprocally altruistic individual is willing to contribute to the pub- lic good if sufficient numbers of other group members are also willing to contribute. Unfortunately, the presence of only a small number of defectors quickly causes coopera- tion to unravel if it is solely based on condi- tionally cooperative behaviour, because the defectors induce the conditional coopera- tors to defect as well. Theory and simulations suggest that reciprocally altruistic strategies can only sustain high levels of cooperation in two-person interactions 7 . Moreover, experi- mental evidence indicates that cooperation in public-good games typically unravels because it is not possible to discipline ‘free riders’ — those who take advantage of others’ cooperation — if only conditionally cooperative strategies are available 8 . news and views NATURE | VOL 432 | 25 NOVEMBER 2004 | www.nature.com/nature 449 In contrast to reciprocal altruism, the notion of altruistic punishment has been more successful in explaining collective action, because direct punishment disci- plines free riders 8 . Altruistic punishers con- tribute to collective actions and are willing to sanction individual defectors even if they incur net costs by doing so. However, within- group selection in the presence of altruistic punishers favours cooperative individuals who do not punish defectors. Such individu- als will never be punished — because they contribute to the collective action — but they also never bear the cost of punishing defectors. These pure cooperators are thus ‘second-order’ free riders because they do not contribute to the disciplining of the self- ish individuals. Therefore, pure cooperators will crowd out altruistic punishers unless there is group competition that renders groups with a higher share of altruistic punishers more successful 9 . Panchanathan and Boyd’s contribution 2 solves this second-order free-rider problem Don’t lose your reputation Ernst Fehr Collective action in large groups whose members are genetically unrelated is a distinguishing feature of the human species. Individual reputations may be a key to a satisfactory evolutionary explanation. Figure 1 Call to arms. Why do humans cooperate with others who are not genetically related to them, particularly in large-scale activities such as the waging of war? Panchanathan and Boyd 2 suggest that each individual is motivated by the desire to maintain their reputation as a contributor to the public good. CHARLES & JOSETTE LENARS/CORBIS ©2004 Nature Publishing Group