Chest X-ray (CR) images are crucial for diagnosing chest lesions. To overcome the issue of insufficiently extracting the relationship between disease features and disease dependency in multi-label disease classification tasks, this paper proposes a chest X-ray multi-label disease classification model based on Vision Transformer (CDCViT). Firstly, Efficientnet-B0 is used as the feature extractor to extract the feature map. After mapping the feature maps, patch embedding and position embedding are added and input into the Transformer module. The Transformer network calculates the weight matrix between features to better mine the relationships between disease features. Then, through Mutual Attention Weight Selection (MAWS), feature selection is performed on the feature tokens collected by multiple Encoder modules, selecting the features most conducive to classification. Finally, the classification results are mapped through a fully connected network. In addition, the ASL loss function is used to calculate the differences between the labels for backpropagation to optimize the model parameters. The proposed model is applied to the public dataset ChestX-ray14. The experimental results show that the CDC-ViT model achieves an average AUC of 0.822 8 for 14 chest diseases, which is about 2% higher than the comparison model, indicating that the proposed CDC-ViT model is superior to many existing classification models.