The purpose of this study was to systematically determine whether low-functioning autistic children [MA of approximately half of their CA or less] could learn through observation by the use of a peer modeling procedure. Since modeling is less structured than traditional one-on-one procedures, it was also thought that modeling might facilitate subsequent generalization of tasks learned through observation. Four autistic children were taught two receptive labeling tasks. One task was taught by a traditional trial-and-error procedure, while the other task was taught by a modeling procedure wherein the models were other autistic children. Results indicated that all four children learned through observation of their peer model. Additionally, generalization and maintenance of correct responding were superior when the children learned through observation rather than by trial and error. These results are discussed in terms of the modeling literature, generalization issues, and implications for designing teaching settings for autistic children.