In several cercopithecine species males exhibit a specific type of male–infant–male interaction during which two males briefly manipulate an infant. These interactions typically occur after a male carrying an infant (infant holder) approaches or is approached by another male who is not holding an infant (infant nonholder). The agonistic buffering and relationship management hypotheses explain these interactions as a tool to establish and maintain social bonds among males. Both hypotheses predict that males preferentially use the opportunity to interact and bond with males dominant to themselves. However, the agonistic buffering hypothesis predicts that males preferentially initiate male–infant–male interactions with the highest ranking males available, whereas the relationships management hypothesis predicts that males are more likely to interact with males that are close to them in rank. To test these predictions, we collected data on 1562 male–infant–male interactions during 1430 hours of focal observation of 12 infants in one group of wild Barbary macaques (Macaca sylvanus) in Morocco. Using generalized linear mixed-effect models we found that males preferably initiated interactions with males that were dominant to them. However, we observed this effect only for interactions initiated by the infant holder. In interactions initiated by non-holders, the receiver’s relative rank did not predict the frequency of interactions. Males also initiated more interactions with males close in rank to themselves than distantly ranked males. Our results support the relationship management hypothesis, but also indicate that the different types of male–infant–male interactions may require different explanations.