This study investigated the impacts of speech (SP) and direct gaze (DG) on attention to faces in 22‐month‐old toddlers with autism spectrum disorder (ASD) and typically developing controls (TD) using the Selective Social Attention task. Toddlers with ASD looked less than TD toddlers at face and mouth regions only when the actress was speaking. Toddlers with ASD looked less at the eye region only when both gaze and speech cues were present. Salience of the combined DG and SP cues was associated concurrently and prospectively with the severity of autism symptoms, and the association remained significant after controlling for verbal and nonverbal levels. The study links poor attention to faces with limited salience of audiovisual speech and provides no support for the face avoidance hypothesis in the early stages of ASD.