Abstract: To solve the problem of feature selection based on tumor gene expression data,this paper presents a feature selection method called LLE Score. The algorithm is a typical filter feature selection method based on the label information of sample. To achieve good feature selection,LLE Score can evaluate the feature vector according to the capability of saving the local neighborhood,and the feature is selected according to the evaluation results. Tumor gene expressive dataset is used by LLE Score to demonstrate the feasibility. The effectiveness of LLE Score is evaluated by the classification accuracy of support vector machine( SVM) classifier.