One of the most common and deadly forms of cancer worldwide is skin cancer. Exposure of the skin to ultraviolet (UV) rays from direct sunlight is the primary cause of this type of cancer. For early-stage diseases to be successfully treated, early and precise identification is essential. In this paper, we propose a technique that extracts global features from an efficient ViT-16 model by turning the image into a sequence of $16 \times 16$ patch tokens, supplemented by positional embeddings. These tokens pass through twelve Transformer Encoders using Multi-Head Self-Attention to produce the final 768-dimensional feature vector which will be used for the final classification task. In order to test the ViT-based classifier, a challenging skin lesion dataset with seven classes was used which is called HAM10000 database. The dataset consists of 10,015 images and it is highly imbalanced. The proposed lightweight model was trained on 80\% of data and tested on the remaining 20\% with 100 epochs utilized during training stage. It achieved a high classification accuracy of 92.62% which outperforms state-of-the-art methods applied on the same dataset.