The giant panda (panda) has lived on the earth for at least eight million years, and as an endangered species, it has received extensive attention from scholars from all walks of life. As an important part of the panda population investigation, the individual identification of pandas can not only provide useful indications but also verify the effectiveness of protection measures. Some work has introduced image processing techniques and deep learning techniques to help researchers identify pandas using face images of a panda. In this paper, we proposed a grouping feature learning method for panda face recognition. In particular, we designed a feature mapping module which can be easily embedded into the existing feature extractors, and a grouping loss function is adopted to constrain the feature mapping, allows the learned similar features to be aggregated together and increase generalization. We use the open captive panda dataset to verify our method, and the results show that our method is effective.