A Bayesian three-level meta-analysis was conducted to examine the effectiveness of Augmented Reality (AR) based interventions in enhancing the domain-specific skills of individuals with autism spectrum disorder (ASD) and/or intellectual and developmental disabilities (IDD). Five moderators, age, gender, disorder type, AR content and skill domains, were further tested to explore the sources of heterogeneity in intervention effectiveness. A total of 28 single-case experimental design studies with 104 individuals were included in the meta-analysis. The results indicate that AR-based interventions have a positive immediate effect on improving skills and that the intervention becomes increasingly effective over time. Although no strong evidence was found for the five moderators examined, patterns emerged indicating that AR interventions had the greatest immediate impact on vocational skills, followed by improvements in physical, living, academic and social skills. Additionally, females and individuals with ASD appeared to benefit more from these interventions. Overall, this study sheds light on the overall AR-based interventions for individuals with ASD/IDD, as well as the potential moderator effects that contribute to the heterogeneity in intervention effectiveness.