Texture analysis is a powerful image analysis method to assess the heterogeneous distribution of tumor quantitatively. Different statistical models have been applied in texture analysis to classify glioma grade and level. It has not been evaluated that which model is the most efficient. The aim of this study is to compare four texture models in glioma grading. Texture features were extracted from multimodality MR images in 3D ROIs. After machine learning and leave-one-out cross validation, the gray-level run-length matrix was found as the best model while gray-level was set as 256.