Traditional Uygur text classification algorithms have disadvantages such as low accuracy and a long operation time. We therefore propose a Uyghur text messages classification method using the deep learning model. Deep learning simulates the multi-layered structure of the brain which gradually extracts data features from low level to high level, and deeply exploits the distribution law of data sets to improve classification accuracy. We use the layered unsupervised method to initialize the deep belief network, and combining with the softmax regression classifier, we realize the classification of Uyghur message data sets. Experiments on Uyghur messages datasets show that compared with the KNN, SVM and the decision tree algorithm, the proposed method has better classification effect.