Ancient paintings are invaluable treasures that offer deep insight into the culture, history, and artistic expression ofcivilizations. Over time, these artworks undergo physical degradation, including the formation of cracks and flakes.Traditionally, such defects are inspected and restored by skilled conservators, but this manual approach is time-consuming,subjective, and may pose risks to the artifacts. With recent advancements in artificial intelligence and deep learning,automated crack detection systems offer promising alternatives for efficient, accurate, and non-invasive analysis. Thisresearch paper presents a deep learning-based approach for detecting surface cracks in ancient paintings. A dataset of highresolution cracked and non-cracked images was prepared and enhanced using augmentation techniques. The study evaluatesmultiple deep learning architectures, including InceptionV3, ResNet50, VGG16, and a custom-built CNN, using transferlearning for improved performance. The model was trained and tested using robust preprocessing and evaluation metrics,achieving high accuracy in binary classification tasks. This research contributes to digital preservation by enabling accuratedetection of defects, thereby assisting conservators in restoration work.Keywords: Ancient paintings, Crack detection, Deep learning, CNN, Digital restoration