Body condition score or body condition assessment, which is a quantitative analysis of the fat content in an animal's body. The transformer model is widely used in computer vision, transformer does not need to rely on the structure of CNN, can achieve good results on image classification tasks, and is suitable for transfer learning. In this paper, we introduce the transformer structure into the field of body condition score for dairy cows. We collected 918 images for scoring the body condition of the dairy cows. On the collected data sets, we use traditional CNN, including ResNext-50, ResNet-50, ECA_NFNet, EfficientNet, EfficientNetV2, ResNext101, HRNet, to score cow body condition, and compare their accuracy with Swin transformer and Vision Transformer. Vision Transformer has achieved good estimation results in comparison with other traditional CNN in the area. Overall accuracy of BCS estimations within 0.25 units of difference from true values was 97.842%, while overall accuracy within 0.50 units was 99.640%. The transformer model for body condition score of dairy cows will be used in the farms gradually.