Despite decades of research, connecting molecular and cellular phenotypes to complex behavioral traits remains an elusive goal(1). Social motivation exhibits individual trait variation(2), which we hypothesize is mediated by molecular and cellular variability across hypothalamic neurons. To test this, we generated single-nucleus RNA-sequencing profiles(3,4) of >120,000 neurons from tuberal hypothalamus and adjacent thalamus in 36 mice, balanced across sex and autism-associated mutation(5), with all mice assessed for social motivation(2). First, we show that molecular activation patterns predict behavior across individuals: specifically, activation of paraventricular Agtr1a+ (angiotensin receptor 1a) neurons predicted reduced social behavior. Subsequent inhibition of AGTR1A with telmisartan-an FDA-approved antihypertensive(6)-improved social orienting. Second, we show natural variation in neuronal proportions-likely arising from stochastic developmental events(7)-is sufficient to shape adult behavior even among genetically-identical individuals: we identified multiple neuronal populations whose relative abundance predicted social reward-seeking behavior. Chemogenetic inhibition of one such population, Nxph4+ neurons of the postero-lateral hypothalamus(8), suppressed multiple aspects of social motivation. This work establishes proof-of-principle for an approach where single-cell genomics precisely maps neural substrates governing behavior. This approach revealed that stochastic variations in neuronal architecture deterministically influence social motivation, and enabled identification of therapeutically-actionable targets with immediate translational potential for disorders with social deficits.