As the development of the medical technology, more and more people start to pay attention to their health. A large amount of health information can be easily obtained from the website now. Therefore, text categorization is important to analyze the information. In this work, we propose a system for text categorization that is based on a Gaussian process. Our proposed system involves the two parts- feature learning and classification. In the first part, we apply the latent Dirichlet allocation (LDA) to obtain the K latent topics proportion from each document. The K-dimensional vector is regarded as the feature of each document. In the classification part, a Gaussian process (GP) is utilized for the text categorization. 10 classes of text documents are categorized by the one-versus-one approach. The experimental results show that our proposed system performs well in text categorization, especially with the small size of training dataset.