Brain arginine vasopressin (AVP) critically regulates normative social behavior in mammals, and experimental disruption of the AVP signaling pathway produces social impairments in rodent models. We therefore hypothesized that AVP signaling deficits may contribute to social impairments in children with autism spectrum disorder (ASD). Since blood measures (which are far easier to obtain than brain measures) of AVP are most meaningful if they are related to brain AVP activity, Study 1 tested the relationship between AVP concentrations in concomitantly collected blood and CSF samples from children and adults (N = 28) undergoing clinical procedures. Study 2 tested whether blood AVP concentrations: 1) differed between children with ASD (N = 57), their ASD discordant siblings (N = 47), and neurotypical controls (N = 55); and 2) predicted social functioning (using the NEPSY-II Theory of Mind and Affect Recognition tasks and the Social Responsiveness Scale) in this large, well-characterized child cohort. Blood AVP concentrations significantly and positively predicted CSF AVP concentrations (F 1,26 = 7.17, r = 0.46, p = 0.0127) in Study 1. In Study 2, blood AVP concentrations did not differ between groups or by sex, but significantly and positively predicted Theory of Mind performance, specifically in children with ASD, but not in non-ASD children (F 1,144 = 5.83, p = 0.017). Blood AVP concentrations can be used: 1) as a surrogate for brain AVP activity in humans; and 2) as a robust biomarker of theory of mind ability in children with ASD. These findings also suggest that AVP biology may be a promising therapeutic target by which to improve social cognition in individuals with ASD.