Purpose <p>Minimally verbal children with autism spectrum disorder (ASD) face profound challenges in social-emotional communication, but objective methods to quantify their facial expression patterns remain limited. This study applied an automated multi-view facial expression recognition (mFER) approach to examine the frequency, synchrony, and social directedness of facial expressions in minimally verbal children with ASD.</p> Methods <p>Fifty-two children aged 3–6 years participated: 18 with ASD, 17 with global developmental delay (GDD), and 17 typically developing (TD) peers. Children engaged in the Smart Observation for Social Communication (SOSC), a series of semi-structured, interaction-based activities adapted from the ADOS-2. Four synchronized cameras captured facial expressions, which were analyzed using a multi-view computer vision analysis.</p> Results <p>Compared with TD peers, the ASD group showed a higher probability of Neutral and Sad expressions and a lower probability of Happiness. These group differences were context-specific, with reduced positive affect during playful interactive activities (e.g., Peekaboo) and increased Sadness during task-based activities (e.g., Puzzle). Children with ASD showed lower facial expression synchrony with the examiner and fewer socially directed expressions. Facial expression features were significantly associated with standardized measures of adaptive functioning and autism symptom severity.</p> Conclusion <p>Integrating the Smart Observation for Social Communication system with multi-view facial expression analysis revealed that children with ASD exhibit context-specific increases in neutral and sad facial expression alongside reduced happiness, synchrony, and socially directed expressions. These clinically valuable features correlated significantly with standardized diagnostic scores, demonstrating the framework’s potential for objective diagnostic complement and outcome monitoring in minimally verbal autism.</p>