The dissertation investigates the contagion of prosocial and antisocial behavior among strangers. We distinguish between two contagion mechanisms: generalized reciprocity (a recipient of social behavior is more likely to pay it forward) and third-party influence (an observer of social behavior is more likely to emulate it). Using two large-scale online experiments, we find that individuals who have benefitted from generosity and suffered from mean behavior are more likely to help and, respectively, harm others. Individuals who observe many acts of kindness are more generous towards others, while individuals who observe few mean acts are less likely to be mean towards others. We then conduct computational experiments with an agent-based model to investigate when the spread of prosocial behavior can become self-sustaining. The results offer explanation for the fact that cultures of kindness are rare for anonymous face-to-face interactions but common online, for example in the form of user-generated content communities.