Skin problems are not only detrimental to physical health but also cause psychological. Especially for patients with damaged or even disfigured faces. In recent years, the incidence of skin diseases has increased rapidly. The medical examination of skin lesions is not a simple task. There are similarities among skin lesions where the doctor’s expe- AQ1 rience with a little inattention can give an inaccurate diagnosis. The automatic classification of skin lesions is expected to save effort, time, and human life. This work has deployed a method using the pre-trained MobileNet model on about 1,280,000 images from the 2014 ImageNet challenge and refined over 25,331 images of the International Skin Imaging Collaboration (ISIC) 2019 dataset. Transfer learning was applied, replacing the classifier with an active softmax layer with three or eight types of skin lesions. An accuracy measure is used to evaluate the performance of the proposed method.