Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by impaired social communication, repetitive behaviors, and restricted interests. Signs of autism usually appear by age 3, but the cinical diagnosis is time-consuming and subjective. Therefore, a rapid and cost-effective assessment method is urgently needed. Children with ASD have atypical gaze patterns when they precept emotional stimuli, which suggests a great potential use of eye-tracking technology as an assessment method for ASD detection. This paper proposes a model for automatically assessing children with ASD based on atypical gaze patterns. The model extracts the eye movement trajectory features of perceived emotions in natural scenes, and uses the machine learning model to learn to automatically identify ASD according to the eye movement trajectory features. Results show that the accuracy reaches 79.71%, which has potentially become an early ASD children screening approach.