[Objectives] The Przewalski’s Horse Equus ferus przewalskii is a nationally protected species in China. The reintroduction projects in the Kalamaili National Park have become a successful example of restoring endangered species. Precise individual identification of Przewalski’s horses is crucial for the appropriate design of conservation and management strategies. Advances in computer vision provide an opportunity for the development of individual identification, with a high degree of accuracy in characteristic recognition. [Methods] In this study, we collected 31 477 images of 20 Przewalski’s Horses in the Kalamaili National Park and used convolutional neural network of deep learning to develop a new individual identification model (Fig. 2). We employed the YOLOv8 model to detect the head images of Przewalski’s Horses and segmented the head images as regions of interest (ROIs). We then used EfficientNetV2 to extract image features from the ROI dataset and identify the individuals (Fig. 3). Furthermore, we used VGG19 and YOLOv8 to extract image features and calculate the identification accuracy, precision, recall, F1 score, and confusion matrix (Figs. 4, 5). [Results] To validate the model performance, we compared the EfficientNetV2 model with VGG19 and YOLOv8 under the same conditions. The EfficientNetV2 model achieves the accuracy of 96.44%, precision of 94.81%, and recall of 94.57%. The VGG19 model has the accuracy of 89.51%, precision of 90.96%, and recall of 89.51%. YOLOv8 reaches the accuracy of 89.33%, precision of 83.90%, and recall of 83.52% (Table 1). The results indicate that the EfficientNetV2 model has the highest accuracy, demonstrating excellent individual identification performance. [Conclusion] This study focused on 20 Przewalski’s Horses in the Kalamaili National Park and developed an individual identification model for Przewalski’s Horses. Our model can be applied in future studies of Przewalski’s Horses, such as accurate individual identification, long-term monitoring, and behavior analysis.