Abstract: The shortages existed in the e-mail feature selection method is first analyzed. A new spam filtering feature selection model based on game theory is then proposed. The game theory is applied to feature selection of mail in order to reduce the scale of information and improve the efficiency of spam filtering. When designing the feature selection model, the impact acted by fuzzy membership of mail samples on feature selection is considered. The feature selection model’s handling capacity for practical problems is enhanced by using a blending sample measure of fuzzy membership function in the definition of feature points to mail category discrimination. The experiments performed on CDSCE Corpus show that the mail feature selection is better than other feature selection methods.