Third-party punishment (TPP) typically happens when uninvolved bystanders sacrifice their self-interests to punish social norm violators. Many studies have found convergent evidence that TPP serves as the fundamental mechanism to enable human with genetic heterogeneity to engage in intense cooperation, where individuals would otherwise be tempted to cheat. So far, however, how TPP has actually evolved and its underpinning cognitive mechanisms are still unresolved. Nature selection theory could hardly explain why individuals as third-parties would punish violators because it reduces their own fitness. While indirect reciprocity theory deals with interpersonal dynamics. And recent studies based on it empirically demonstrated that third-party punishers have higher levels of within-group fitness in two ways. First, TPP is a costly signal of trustworthiness in the partner-choice model. Third-party punishers are trusted more by observers than those who do not punish norm violators. Meanwhile, they are indeed more trustworthy than non-punishers. Second, TPP also functions as a deterrence to cheaters in the partner-control model. Third-parties are driven to punish norm violators in order to protect their self-interests by their inference about how norm violators will treat them in the future, and norm violators may be more deterred by third-party punishers than by non-punishers. The exact function of TPP is dependent on the specific context where TPP happens. In particular, if TPP happens in the context of large-scale interpersonal interaction, it functions as a signal of trustworthiness, and as a signal of deterrence in the context of small-scale interpersonal interaction. According to our literature review, we found that TPP is a complex altruistic behavior involving multi-level systems, including emotional systems like guilt, socio-cognitive systems like empathy, and central executive systems like cognitive control. The relationship between empathy and TPP is not as simple as we commonly assumed, i.e., higher empathy is associated with more TPP, but is dependent on context. People with higher levels of empathy tend to help victims if they have the choice to help; instead, if to punish violators is the only choice to enforce social norms, they are inclined to do so. This illustrates how TPP is a special kind of altruistic behaviors. Moreover, cognitive control plays an important role in imposing TPP. First, with the help of cognitive control to inhibit selfishness, individuals as third-parties would be able to resolve the cognitive conflict between maximizing self-interests and sacrificing one’s own interest to enforce norms. Second, cognitive control might modulate the relationship between empathy and TPP, particularly in regulating personal distress, a dimension of empathy negatively correlated with altruism. It is expected to be of great help to understand and solve the practical problems of social governance if we find more what contributes to TPP. Considering TPP is a special kind of altruistic behaviors, future studies could investigate how TPP is different from other kind of altruistic behaviors such as help. Another issue is that the underlying motivations of TPP evoked by guilt and anger may be different, and thus future studies could employ various techniques like eye-tracking, EEG, fMRI, to differentiate them in order to make contribution to the understanding of TPP.