Background More refined dimensions of social‐communication impairment are needed to elucidate the clinical and biological boundaries of autism spectrum disorders (ASD) and other childhood onset psychiatric disorders associated with social difficulties, as well as to facilitate investigations in treatment and long‐term outcomes of these disorders. Methods This study was intended to identify separable dimensions of clinician‐observed social‐communication impairments by examining scores on a widely used autism diagnostic instrument. Participants included verbally fluent children ages 3–13 years, who were given a clinical diagnosis of ASD (n = 120) or non‐ASD (i.e. ADHD, language disorder, intellectual disability, mood or anxiety disorder; n = 118) following a comprehensive diagnostic assessment. Exploratory and confirmatory factor analysis examined the factor structure of algorithm items from the Autism Diagnostic Observation Schedule (ADOS), Module 3. Results Results indicated that a three‐factor model consisting of repetitive behaviors and two separate social‐communication behavior factors had superior fit compared to a two‐factor model that included repetitive behaviors and one social‐communication behavior factor. In the three‐factor model, impairments in ‘Basic Social‐Communication’ behaviors (e.g. eye contact, facial expressions, gestures) were separated from impairments in ‘Interaction quality.’ Confirmatory factor analysis in an independent sample of children in the Simons Simplex Collection (SSC) further supported the division of social‐communication impairments into these two factors. Scores in Interaction Quality were significantly associated with nonverbal IQ and male sex in the ASD group, and with age in the non‐ASD group, while scores in basic social communication were not significantly associated with any of these child characteristics in either diagnostic group. Conclusions Efforts to conceptualize level, or severity, of social‐communication impairment in children with neurodevelopmental disorders might be facilitated by separating the most basic (or proximal) social‐communication impairments from those that could arise from a range of other phenotypic variables. Identification of social‐communication subdimensions also highlights potential avenues for measuring different types of social‐communication impairments for different purposes (e.g. for differential diagnosis vs. response to treatment).