PURPOSE: Children with autism spectrum disorder (ASD) have been found to exhibit lower quantities of pretend play than do typically-developing children, which could be related to their symptom severity. However, how symptom severity is associated with pretend play in children with ASD is unknown. Therefore, we applied an elaborate pretend play assessment to characterize the pretend play of children with ASD with different symptom severity levels. DESIGN: This study used a quasi-experimental design. Children were included if they were at a chronological age of 4-12 years old and had received a formal diagnosis of ASD. Children were excluded if they (1) had symptoms associated with organic brain dysfunction or chromosomal abnormality, (2) had uncorrected hearing or visual impairments, and (3) were unable to follow orders or complete the procedures. METHOD: Children's pretend play performance, autistic behaviors, and verbal comprehension were respectively assessed with the Child-Initiated Pretend Play Assessment (ChIPPA), Childhood Autism Rating Scale (CARS), and Verbal Comprehension Index of the Wechsler Intelligence Scale. The ChIPPA includes three types of raw scores: number of imitated actions (NIA), number of object substitutions (NOS), and percentage of elaborate pretend play actions (PEPA). The three types of scores combined with two contexts of conventional and symbolic play generate 6 raw scores. Children were categorized according to their CARS cut point scores into 3 severity levels: non-ASD, mild/moderate, and severe. One-way analysis of variance (ANOVA) and Tukey Post hoc analysis were applied to examine if the three severities were associated with differences in performance in pretend play. RESULTS: A total of 72 children with ASD aged 70-141 months were enrolled. The ANOVA analysis revealed that the 3 severity groups had significant differences symbolic PEPA (F.sub.(2,65) = 3.8, p < 0.05), symbolic NOS (F.sub.(2,65) = 3.2, p < 0.05), and total PEPA scores (F.sub.(2,65) = 3.6, p < 0.05). The post hoc results indicated significant differences between non-ASD and severe ASD groups in symbolic PEPA (group difference = 15.7, p < 0.05), symbolic NOS (group difference = 3.9, p < 0.05), and total PEPA scores (group difference = 27.9, p < 0.05). The results demonstrate that children without ASD display higher frequencies of symbolic object substitution and higher percentages of elaborate pretend play actions than do children with severe ASD during pretend play. CONCLUSION: Our findings demonstrate that symptom severity is a significant variable of elaborateness and frequency in symbolic play. Children with very mild autistic traits have more complexity, creativity, and flexibility in symbolic play than do children with severe autistic traits. This study provides evidence on the association of symptom severity with pretend play performance in children with ASD for both clinical and research settings.