Naturalistic observations of infant/caregiver social attention have yielded rich information about human social development. However, observational data are expensive, laborious, and reliant on fallible human coders. We model interactions between caregivers and infants using a three dimensional simulation environment in order to gain greater insight into the development of infant attention sharing, specifically gaze following. Most models of infant cognition have been only abstractly linked to the detail of a real life environment and to the perception-and-action physicality of human infants. Our simulation uses human data from videotaped infant/caregiver interactions and a rich 3D environment to model the development of gaze following. Initial tests suggest that infant gaze following can be learned in our simulation using parameters derived from behavioral data.