Glaucoma, a group of eye diseases, damages individual eye health by injuring the optic nerve, and this leads to inevitable vision loss. Although the symptoms of glaucoma can be observed by experts, the procedure for doing so is still complex and time-consuming. This problem has become more acute in China than in other locations due to its high population and limited medical resources. With the development of IT and health informatics, automatic diagnosis has been found to be effective for managing the diagnosis issues with regard to glaucoma. However, one crucial yet underexplored problem is how to improve the effectiveness of automatic diagnosis by considering uncertainty and gathering key information from multimodal data, including medical indicators, images, and texts. Therefore, this study proposes a Bayesian deep multisource learning (BDMSL) model to address these problems. Specifically, multisource learning is introduced to integrate data from multiple sources, while Bayesian deep learning is adopted to obtain model uncertainty information. Based on real medical data collected from one of the best eye hospitals in China, our research results show that the BDMSL model achieves better performance than other methods in terms of glaucoma detection. With the exploration of health informatics in terms of diagnosing glaucoma in China, the proposed model can be generalized to provide health services globally.