Autism is a neurodevelopmental disorder that often manifests in childhood, characterized by social difficulties and repetitive behaviors. Motor impairments are also common, with autistic children experiencing challenges in converting sequential actions into integrated movements, affecting fine motor skills and daily activities like handwriting. This study explores the handwriting characteristics of autistic children through an analysis of Chinese characters. Our research has two objectives. First, we identify the handwriting characteristics of autistic children and typically-developing children through an analysis of Chinese characters. Second, we establish neatness criteria to assess whether a Chinese character is written neatly, aiming to understand handwriting neatness in both groups and to train a classification model using only neatly written characters. The dataset is derived from elementary school workbooks, reflecting real-life situations. To analyze handwriting features, we employ various machine learning/deep learning models equipped with the Class Activation Map (CAM) technique to provide the interpretability of the models by the visualization of the models’ focal points. By applying oversampling techniques for data balancing, our model achieves an F1-score of 0.9720, surpassing previous studies on classifying autistic children and typically-developing children. Our findings contribute to a deeper understanding of autistic children’s handwriting characteristics, providing insights that can support early detection and intervention. The models can therefore serve as a valuable tool for educators and parents to assess the children’s autistic tendencies and support their handwriting development.