The Gilliam Autism Rating Scales-3rd Edition (GARS-3) is widely utilized, but evidence regarding its psychometric properties is limited and equivocal. The current study attempted to create a shorter GARS-3, to be utilized for screening purposes, via exploratory factor analysis. Archival data from 183 clinically referred children ( \(\overline{X }\) age = 9.03; autism spectrum [AUT], n = 86; Not-Autism Spectrum [NOT], n = 97) who received comprehensive assessments with the Autism Diagnostic Observation Schedule-2nd Edition (ADOS-2) were analyzed. Retention of items with factor loadings > .70 resulted in a 21-item (∝ = .92), 5-factor measure (SF21) that was invariant across AUT/NOT groups to the scalar level. Evidence for discriminant validity with the Child Behavior Checklist was moderate; evidence for concurrent and criterion validity was weak to poor. Similar to multiple prior studies, sensitivity and specificity could not be optimized at a given cutoff. Neither the GARS-3 nor the GARS-3SF21 is recommended for autism diagnosis in complex clinically referred samples. Patterns of findings that could be used to improve autism questionnaires are identified and discussed.
