With the rapid growth of the SMS, the filtration to all messages has been unable to meet the real-time processing requirement. In this paper, we propose a sampling of mass SMS filtering algorithm based on frequent time-domain area to solve this problem. First, we collect the long-running system log. And then analyze the time and domain features of the messages to generate the time-domain strategy. Finally we predict the potential spam messages’ rate in different domain and different time, and carries on the filtration according to each rate separately. This algorithm can satisfy the real-time filtration requirement of the mass SMS stream, and meanwhile there is no significant reduction in spam.