Little is understood about how children attend to and learn from gaze when learning new words, and whether gaze confers any benefits beyond word mapping. We examine whether 6- to 11-year-old typically-developing children (n = 43) and children with Autism Spectrum Disorder (n = 25) attend to and learn with gaze differently from another directional cue, an arrow cue. An eye-tracker recorded children’s attention to videos while they were taught novel words with a gaze cue or an arrow cue. Videos included objects when they were static or when they were manipulated to demonstrate the object’s function. Word learning was measured immediately after videos and one week later. In contrast to an arrow cue, children in both groups looked longer at a gaze cue and had more contingent looks from the gaze cue to the referent. Exploratory analyses demonstrated that across both groups, children with higher versus lower social competence skills recalled more semantic features about the referent in the gaze condition. We discuss how these findings add to our theoretical understanding of how gaze supports word learning.