Autism spectrum disorder (ASD) refers to a neurodevelopmental disorder that is described as having issues with socialization, communication, as well as repetitive behavior. Early detection of ASD is critical at an early age so that individuals can receive the required intervention and better developmental outcomes. The conventional methods of diagnosis are generally long procedures which are subjective and need the assessment of experts. This paper presents a solution to the limitations in order to come up with a deep learning model that will be used in detecting autism on the basis of the facial images automatically. This study implements a Convolutional Neural Network (CNN) model that categorizes the images of faces as autistic and non-autistic. Resizing and normalization is a process that is involved in pre- processing so as to create homogenous data. The CNN model suggested here automatically extracts the relevant features of the face and generate accurate diagnostic classifications. The measures of performance are accuracy, precision, recall and F1-score. The model has been demonstrated to be accurate in its operation by both experimentation and is seemingly a comprehensive all inclusive, non-invasive and helpful instrument of early childhood autism diagnosis.