The Autism Diagnostic Observation Schedule (ADOS) is widely used to assess symptoms of autism spectrum disorder (ASD). Given well-documented differences in social behaviors across cultures, this study examined whether item-level biases exist in ADOS scores across sociodemographic groups (race, ethnicity, and gender). We examined a subset of ten ADOS items among participants (N = 2458). Holding level of overall ADOS behavioral symptoms constant, we found significant item level bias (measurement noninvariance) for race and ethnicity on three ADOS items. Item-level bias was not apparent across gender. Although the magnitude of bias was small, our findings highlight the need to reevaluate norms and operational definitions used in assessments to increase ASD diagnostic accuracy among culturally-diverse groups.