Five participants with autism spectrum disorder (ASD) between the ages of 10 and 14 years attended a twice-weekly social skills group for 8 weeks. A multiple probe design across skills with concurrent replication across participants was utilized to assess the effects of social skills training using multiple exemplars and lag schedules on social skill variability. Prior to intervention, participants demonstrated low levels of skill accuracy and appropriate variability. Training with one and three exemplars of target skills without lag schedules resulted in no to small increases in appropriate variability, despite increases in skill accuracy. Following introduction of lag schedules, substantial increases were observed. Post-intervention rating scales completed by parents of participants indicated improvements in social functioning and reductions in repetitive behavior.