The paper, aimmed at spam filter, at first separationing, preproccessing and building text vector for the obtained spam mails and legitimate mails, then proccessing vector dimensional reduction using four common key extraction methods, and based on this, presents a comprehensive key extraction algorithm, which takes front n key words of their intersection as a candidate word for classification test according to sort results of each assessment function. Finally, Simulation verifies the effection of “n” on the classification in the algorithm, thus verifying the effectiveness of the proposed algorithm.