Autism disorder is a neurologic illness which impairs verbal abilities and social interaction in children aged 6 to 17. It can be detected by using facial or biomedical images. As Autistic children has not similar facial features as normal child, Deep learning techniques may be used by extracting some important features. The key purpose of this research paper is to design a system to detect this order using facial features. Predefined models like VGG 16, VGG19, ResNet152V2, EfficientNetV2 and InceptionResNetV2 have been applied on Dataset Autistic Children Facial Dataset is taken from Kaggle data owned by Imran Khan. Different evaluation parameters like accuracy, loss, Confusion matrix are used for the evaluation of these five predefined models. It has been found that ResNet152V2 has achieved higher accuracy on other models.