Data privacy protection is an important research area, which is especially critical in this big data era. To a large extent, the privacy of visual classification tasks is mainly in the one-to-one mapping between image and its corresponding label, since this relation provides a great amount of information and can be used in other scenarios. In this paper, we propose Mapping Distortion Protection (MDP) and its augmentation-based extension (AugMDP) to protect the data privacy by modifying the original dataset. In MDP, the label of the modified image does not match the ground-truth mapping, yet DNNs can still learn the ground-truth relation even when the provided mapping is distorted. As such, this method protects privacy when the dataset is leaked. Extensive experiments are conducted on CIFAR-10 and restricted CASIA-WebFace dataset, which verifies the effectiveness and feasibility of the method.