Aggregating welfare across individuals to reach collective decisions is one of the most fundamental problems in our society. Interpersonal comparison of utility is pivotal and inevitable for welfare aggregation, because if each person's utility is not interpersonally comparable, there is no rational aggregation procedure that simultaneously satisfies even some very mild conditions for validity (Arrow's impossibility theorem). However, scientific methods for interpersonal comparison of utility have thus far not been available. Here, we have developed a method for interpersonal comparison of utility based on brain signals, by measuring the neural activity of participants performing gambling tasks. We found that activity in the medial frontal region was correlated with changes in expected utility, and that, for the same amount of money, the activity evoked was larger for participants with lower household incomes than for those with higher household incomes. Furthermore, we found that the ratio of neural signals from lower-income participants to those of higher-income participants coincided with estimates of their psychological pleasure by "impartial spectators", i.e. disinterested third-party participants satisfying specific conditions. Finally, we derived a decision rule based on aggregated welfare from our experimental data, and confirmed that it was applicable to a distribution problem. These findings suggest that our proposed method for interpersonal comparison of utility enables scientifically reasonable welfare aggregation by escaping from Arrow's impossibility and has implications for the fair distribution of economic goods. Our method can be further applied for evidence-based policy making in nations that use cost-benefit analyses or optimal taxation theory for policy evaluation.