Participant characteristics.

Difficulties with emotion recognition can occur in neurodevelopmental conditions, including in autistic individuals. Providing interventions to support this would therefore be beneficial, particularly in terms of downstream effects on wellbeing, social relationships and education. In this online experimental study, we examined the effect of a recently developed facial emotion recognition training task versus a sham/control task in an adult population who self-identified as autistic over four sessions in a 2-week period, with a fifth follow-up session (N = 184). Our main analyses showed that facial emotion recognition accuracy was greater in Session 4 in the active group, with an estimated improvement of 14% (equivalent to approximately 7 additional correct responses), compared to 2% (equivalent to approximately 1 additional correct responses) in the sham group (p = 4x10-09). Additional analyses suggested training effects were generalisable to facial stimuli that participants had not been trained on and were still present, although attenuated, two weeks later. We also observed some self-reported improvements in social interactions post-training. Overall, this study demonstrated improved emotion recognition with this training task in an adult sample who self-identified as autistic. Future work is needed to investigate the effect of this task on emotion recognition accuracy in those with a formal diagnosis of autism, and in autistic children where support could be most beneficial.