Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by persistent challenges in social communication. Early intervention is paramount; however, traditional diagnostic pathways often take years due to a lack of specialized clinicians. This research proposes an automated screening tool using facial image analysis. By employing the VGG16 architecture via Transfer Learning, we extract high-level spatial features from facial landmarks to identify markers associated with ASD. Our findings indicate that computational models can provide a significant preliminary screening layer, reducing the burden on clinical resources.