["Autism Research, EarlyView. ", "\nABSTRACT\nThe current study investigated 1144 toddlers and preschoolers (ASD + GDD n = 592; ASD only n = 249; GDD only n = 89; no ASD or GDD n = 214) with the toddler module (38.6%), Module 1 (57.5%), and Module 2 (3.9%) as well as Mullen Scales of Early Learning. The calibrated severity score (CSS) was used to compare severity across modules. The study sample was stratified by GDD (Visual Reception Developmental Quotient < 75), and each stratum was investigated with descriptive statistics, ROC curves, and test statistics to identify the optimal cut‐off CSS to differentiate ASD and non‐ASD. ROC analysis indicated that the CSS scores showed excellent discrimination for ASD status for both the GDD (AUC = 0.86) and no GDD (AUC = 0.95) strata. In the no‐GDD stratum, an ADOS‐2 CSS of 5 was determined to be the optimal cut‐off. In the GDD stratum, an ADOS‐2 CSS of 6 was determined to be the optimal cutoff. While non‐spectrum/little‐to‐no concern and autism/moderate‐to‐severe concern showed very high predictive accuracy for diagnostic outcomes, the autism spectrum/mild‐to‐moderate concern lacked clear diagnostic directionality, regardless of GDD status. This is the first study with a large sample of toddlers and preschoolers exploring optimal ADOS‐2 CSS cut‐off when stratified by GDD.\n"]