Purpose: To assess the generalizability of findings from randomized controlled trials (RCTs) evaluating emotion recognition (ER) training for children and adolescents with autism spectrum disorder (ASD). Methods: We present a systematic review and narrative synthesis of the determinants of external validity in RCTs on ER training. Generalizability of the findings across situations, populations, settings, treatment delivery, and intervention formats was considered. Results: We identified 13 eligible studies. Participants were predominantly boys with ASD in the normative IQ range (IQ > 70), with an age span from 4 to 18 years across studies. Interventions and outcome measures were highly variable. Several studies indicated that training may improve ER, but it is still largely unknown to what extent training effects are translated to daily social life. Conclusion: The generalizability of findings from currently available RCTs remains unclear. This underscores the importance of involving children with ASD and their caregivers in informed treatment decisions.