The GARS-2 (Gilliam 2006) is widely used for autism screening; however, the validity of its three conceptually-derived subscales has not been evaluated. In this study, exploratory (EFA) and confirmatory (CFA) factor analyses of data from matched subgroups of the standardization sample did not support the GARS-2 subscale structure. EFAs identified four factors in the model development subgroup (n = 496): "stereotyped/repetitive behavior," "stereotyped/idiosyncratic language," "word use problems," and "social impairment." CFAs supported this model in the replication subgroup (n = 494). Findings suggest that the GARS-2 subscales have limited clinical utility and that the Autism Index should be interpreted with caution. Although more research is needed, the clinical utility of the four empirically-derived scales may be limited by factors related to item content and test development procedures.