Mobile phone is a embedded system with limited energy, memory and computing ability. The traditional SMS spam filtering algorithms only focused on the accuracy rate of spam filters and did not consider the consumption of mobile phone’s hardware resources. This paper proposes an optimized Naïve Bayes SMS Spam filtering algorithm, the main purpose is to reduce the consumption of hardware resources and reflect user’s individual preference. A feature library is designed to save weight of features, conditional probabilities and prior probabilities. The filtering algorithm is divided into a feature library updating algorithm and an online SMS filtering algorithm. The feature library is updated regularly by the feature library updating algorithm. The feature library updating algorithm does not need to run immediately, and can be performed in free time of mobile application or copied to PC client asynchronously. The online SMS filtering algorithm running on a mobile phone performs only the minimum amount of work which does not require much resources. It can be quickly adapted to the changing of user’s preference. Experimental results show that the algorithm can improve the speed of classification while maintaining a high classification accuracy. it takes up less storage space and can be used on common mobile platforms.