Background Discrepancy between informants (parents and teachers) in severity ratings of core symptoms commonly arise when assessing autism spectrum disorder (ASD). Whether such discrepancy yields unique information about the ASD phenotype and its clinical correlates has not been examined. We examined whether degree of discrepancy between parent and teacher ASD symptom ratings defines discrete, clinically meaningful subgroups of youth with ASD using an efficient, cost‐effective procedure. Methods Children with ASD (N = 283; 82% boys; Mage = 10.5 years) were drawn from a specialty ASD clinic. Parents and teachers provided ratings of the three core DSM‐IV‐TR domains of ASD symptoms (communication, social, and perseverative behavior) with the Child and Adolescent Symptom Inventory‐4R (CASI‐4R). External validators included child psychotropic medication status, frequency of ASD‐relevant school‐based services, and the Autism Diagnostic Observation Schedule (ADOS‐2). Results Four distinct subgroups emerged that ranged from large between‐informant discrepancy (informant‐specific) to relative lack of discrepancy (i.e. informant agreement; cross‐situational): Moderate Parent/Low Teacher or Low Parent/Moderate Teacher Severity (Discrepancy), and Moderate or High Symptom Severity (Agreement). Subgroups were highly distinct (mean probability of group assignment = 94%). Relative to Discrepancy subgroups, Agreement subgroups were more likely to receive psychotropic medication, school‐based special education services, and an ADOS‐2 diagnosis. These differential associations would not have been identified based solely on CASI‐4R scores from one informant. Conclusions The degree of parent–teacher discrepancy about ASD symptom severity appears to provide more clinically useful information than reliance on a specific symptom domain or informant, and thus yields an innovative, cost‐effective approach to assessing functional impairment. This conclusion stands in contrast to existing symptom clustering approaches in ASD, which treat within‐informant patterns of symptom severity as generalizable across settings. Within‐child variability in symptom expression across settings may yield uniquely useful information for characterizing the ASD phenotype.