We study the co-evolution of social preferences and bounded rationality. In particular, weshow that when agents are boundedly rational, altruistic preferences are evolutionarilystable, even in environments that are deemed unfavorable for altruism in the literature. Theexisting standard result is that when interactions are strategic substitutes and exhibit negativeexternality, only selfish preferences are evolutionary stable. The key assumption underlyingthis result is that agents are perfectly rational. Selfish agents are thus able to play the Nashequilibrium, gaining evolutionary advantages over altruists. By relaxing this assumption, weshow that altruist preferences can survive among bounded rational agents. The simpleintuition is that selfish agents, now with bounded rationality, choose excessive action, whichin turn induces altruists to choose an action level closer to the Nash equilibrium–an actionlevel evolutionarily stable in the long run. We combine the level-k model of boundedrationality and the standard evolutionary model of altruistic preferences and characterize forthe conditions under which altruism can proliferate in the long run.