OBJECTIVES:Primates vary in their sleep durations and, remarkably, humans sleep the least per 24-hr period of the 30 primates that have been studied. Using phylogenetic methods that quantitatively situate human phenotypes within a broader primate comparative context, we investigated the evolution of human sleep architecture, focusing on: total sleep duration, rapid eye movement (REM) sleep duration, non-rapid eye movement (NREM) sleep duration, and proportion of sleep in REM. MATERIALS AND METHODS:We used two different Bayesian methods: phylogenetic prediction based on phylogenetic generalized least squares and a multistate Onrstein-Uhlenbeck (OU) evolutionary model of random drift and stabilizing selection. RESULTS:Phylogenetic prediction confirmed that humans sleep less than predicted for a primate of our body mass, predation risk, brain size, foraging needs, sexual selection, and diet. These analyses further revealed that humans pack an unexpectedly higher proportion of REM sleep within a shorter overall sleep duration, and do so by reducing NREM sleep (rather than increasing REM). The OU model generally confirmed these findings, with shifts along the human lineage inferred for TST, NREM, and proportion of REM, but not for REM. DISCUSSION:We propose that the risks and opportunity costs of sleep are responsible for shorter sleep durations in humans, with risks arising from terrestrial sleep involving threats from predators and conspecifics, and opportunity costs because time spent sleeping could be used for learning, creating material objects, and socializing.