The current study examines the use of statistical information in the sociomoral evaluation of inconsistent agents. 108 college students participated across three studies. In Study 1, participants read stories about predominantly prosocial and predominantly antisocial agents and then rated the agent’s morality and their preference toward the agent in a 7-point Likert scale. The repeated measures ANOVA analysis revealed that the participants preferred and evaluated the predominantly prosocial agent as more moral than the predominantly antisocial agent. In Study 2, the comparison between consistently prosocial (antisocial) and inconsistently prosocial (antisocial) agents revealed that the participants evaluate an agent’s social-morality based on statistical information. Finally, Study 3 investigates whether frequency or ratio information is prioritized when evaluating an inconsistent agent. Participants preferred and evaluated the predominantly prosocial agent as more moral regardless of the number of actions that constituted each scenario (2:1 or 6:3). The result suggests that the ratio information is prioritized when evaluating the social-morality of an inconsistent agent. In summary, the current study suggests that inconsistent agents’ sociomoral evaluation is based on statistical information.