There are many theories of the function of dreams, such as memory consolidation, emotion processing, threat simulation and social simulation. In general, such theories hold that the function of dreams occurs within sleep; occurs for unrecalled dreams as well as for dream that are recalled on awakening; and that conscious recall of dreams is not necessary for their function to occur. In contrast, we propose that dreams have an effect of enhancing empathy and group bonding when dreams are shared and discussed with others. We propose also that this effect would have occurred in history and pre-history and, as it would have enhanced the cohesiveness and mutual understanding of group members, the fictional and engaging characteristics of dream content would have been selected for during human social evolution, interacting with cultural practices of dream-sharing. Such dream-sharing may have taken advantage of the long REM periods that occur for biological reasons near the end of the night. Dream-production and dream-sharing may have developed alongside story-telling, utilising common neural mechanisms. Dream-sharing hence would have contributed to Human Self-Domestication, held by many researchers to be the primary driver of the evolution of human prosociality, tolerance and reduced intragroup emotional reactivity. We note that within-sleep theories of dream function rely on correlational rather than experimental findings, and have as yet untested and speculative mechanisms, whereas post-sleep effects of dream-sharing are easily testable and have mechanisms congruent with the social processes proposed by the theory of Human Self-Domestication.