Text representation is the most important phase in automatic text categorization.In the Vector Space Model(VSM) based text representation,the selection of feature granularity has the direct impact on the text categorization performance.The statistical approach based Uyghur phrase extraction algorithm was proposed and the Uyghur text categorization experiments was conducted using Support Vector Machine(SVM) algorithm based on the extracted phrases as text features.The experimental results show that the phrase based Uyghur text categorization achieves higher classification precision and recall compared to the word based categorization.