The evolutionary and neural underpinnings of human prosociality are still largely unknown. A growing body of evidence suggests that some species find the sight of another individual receiving a reward reinforcing, often called vicarious reinforcement. One hypothesis is that vicarious reward is reinforcing because it is arousing like a primary reward. We evaluated this hypothesis by measuring the autonomic pupil response of eight monkeys across two laboratories in two different versions of a vicarious reinforcement paradigm. Monkeys were cued as to whether an upcoming reward would be delivered to them, another monkey, or nobody and could accept or decline the offer. As expected, all monkeys in both laboratories showed a marked preference for juice to the self, together with a reliable prosocial preference for juice to a social partner compared to juice to nobody. However, contrary to the autonomic arousal hypothesis, we found that pupils were widest in anticipation of juice to the self, moderately-sized in anticipation of juice to nobody, and narrowest in anticipation of juice to a social partner. This effect was seen across both laboratories and regardless of specific task parameters. The seemingly paradoxical pupil effect can be explained by a model in which pupil size tracks outcome salience, prosocial tendencies track outcome valence, and the relation between salience and valence is U-shaped.