For instance, with extreme changes in climate, weather, and environmental influences, skin diseases have become more and more dangerous and common. Therefore, the identification and detection of skin diseases are important researches to help patients have timely prevention and treatment solutions. In this study, we propose using a transfer learning approach to detect and identify skin diseases. The input image will be preprocessed, segmented, and used transfer learning from the pre-trained VGG19 deep learning model to identify skin diseases. Different scenarios were tested on the set of images collected from the International Skin Imaging Collaboration (ISIC). This dataset contains a total of 2655 images (including male and female patients). The data were classified into 3 categories which are carcinoma disease (1127 images), skin hemorrhagic disease (1006 images), and normal skin (522 images). These data were labelled by experts in the field of dermatology. The experimental results show that using the transfer learning method from the pre-trained VGG19 model is very positive with the accuracy and the F1 measure of 0.85–0.84, 0.93–0.93, and 0.87–0.86, respectively, for 3 test scenarios.