Accurate identification of cattle behavior is crucial for monitoring their health, welfare, and productivity, making it a vital aspect of precision livestock farming. This study explores the application of Transformer models for classifying cow behaviors using time-series data derived from tri-axial accelerometer sensors. By leveraging the Transformer’s capability to capture long-term dependencies and temporal patterns in sequential data, this method seeks to reveal intricate behavioral patterns within cattle movement data. The model is tested on a diverse dataset, encompassing behaviors such as feeding, resting, ruminating, moving, and salting, among others. Evaluation results indicate that the Transformer-based model achieves high levels of accuracy, precision, and recall, affirming its suitability for automated monitoring of cattle behaviors. This approach holds promises for enhancing animal welfare, boosting productivity, streamlining farm management, and improving decision-making in the livestock industry.