Traffic sign classification is an essential challenge for autonomous driving. Camera-based computer vision techniques have been proposed for this purpose, employing several convolutional neural network (CNN) architectures that have been tested and proved using multiple available datasets. Recently, novel transformer-based models have been introduced for diverse computer vision applications, attaining state-of-the-art performance and surpassing CNNs in several areas. In this study, we implement a vision transformer (ViT) model using the German traffic sign (GTSRB) dataset, containing 43 classes. The ViT model is trained and tested using the GTSRB dataset. Furthermore, the model’s performance is evaluated using accuracy, precision, recall, and F1-score. The ViT model is also compared with the pre-trained CNN models, including EffecientNet, InceptionNet, VGG19, and the recently proposed TSC18 CNN. The results show that the ViT model outperforms with 99.65% accuracy, which is higher than the pre-trained and previously proposed CNN models. The ViT model is able to perform better than the CNN models for traffic sign classification.