Children on the autism spectrum often have deficits in language development. Given the resources required to provide intensive intervention services, either in person or via telehealth, it may be important to maximize instructional time by using procedures that lead to generative language use, such as matrix training. Matrix training involves systematically arranging and selecting multi-component instructional targets, in which only a select few of the component combinations are directly taught; then, following mastery of the selected targets, a check is completed to test for the emergence of the rest of the combinations within the matrix. In the current study, we use a nonconcurrent multiple baseline design across participants to determine the effects of telehealth-based matrix training on trained and untrained language targets for three young children on the autism spectrum. Two of the three participants required booster training sessions to reach mastery in the post-test condition. These preliminary results suggest that matrix training may be an effective method for teaching multicomponent tacting skills via telehealth to young children with autism.