The Sundanese script is a cultural heritage of the Sundanese people that deserves to be preserved. With the advances in computer vision and deep learning, the study of Indonesian local character recognition still has room for improvement and therefore needs to be encouraged. In this study, the Sundanese script used was Swara and Ngalagena script. The classification of the Sundanese script poses a significant challenge due to the complexity and variability of the script's visual patterns. Therefore, this study classifies Sundanese script using the deep learning method, namely the Transfer Learning method based on the Convolutional Neural Network. The architectures used in this research are ResNet-50, VGG-19, and MobileNet. Then to evaluate the performance of the three architectures, the evaluation metrics used in this study are accuracy, precision, recall, and F1 scores will be used. This study conducted two experiments that added a layer, namely GlobalPooling2D, for the first experiment, and the second experiment added a Flatten layer. The three CNN architectures show very satisfactory results in classifying Sundanese script. The best accuracy results using the first experiment are achieving an accuracy value of 99% for the VGG-19 architecture. The ResNet-50 architecture is 87%, while the MobileNet architecture produces an accuracy of 60%. Overall, the VGG-19 architecture is the best in this study of Sundanese script classification.