<p>One of the most common pathologies on masonry building facades is mortar loss, which can lead to insufficient bonding force, uneven stress distribution between blocks and reduced structural bearing capacity. Traditional manual visual detection is a laborious, inefficient and subjective process often conducted through sampling inspection, which cannot fully reflect the overall quality of masonry. In contrast, automatic detection based on deep learning can significantly enhance accuracy and efficiency. Therefore, the goal of this paper is to validate the feasibility of applying deep learning methods for the automatic detection of mortar loss in images of masonry facades. To achieve this, datasets for classification and segmentation were created with complex backgrounds and varying degrees of mortar loss, based on field investigations and smartphone photography results. Several convolutional neural networks were compared, revealing MobileNetV3 as the best-performing model at the patch level, achieving an accuracy of 98.68%, and U-Net as the top performer at the pixel level, with an F1 score of 87.2%. Finally, a two-step automatic detection method for mortar loss, combining deep learning models with sliding window algorithm, was proposed. Experimental results demonstrate its capability to accurately and efficiently detect mortar loss in practical scenarios, with strong generalization performance and noise resistance.</p>