Nowadays, spam messages have been overflowing in many countries. They seriously violate personal rights, and may even harm the national security. The existing filtering techniques usually uses traditional text classifiers, which are more suitable to deal with normal long texts, therefore, it often faces some serious challenges, such as the sparse data problem and noise data in the SMS message. This research work proposes a message topic model (MTM) for SMS spam filtering. The MTM derives from the famous probability topic model. Although the MTM is based on probability topic model, it is different from the famous standard Latent Dirichlet Allocation (LDA) in the following aspects: (1) For the purpose of overcoming the sparsity problem in SMS message classification, first, the standard K-means algorithm is used to classify the training data into rough classes, then, aggregates all the spam messages of a class into a single document. (2) Symbol semantics is taken in account. Some preprocessing rules and background terms are considered to make the model more appropriate to fully represent SMS spam. Finally, we compare the MTM with the SVM and the standard LDA on the public SMS spam corpus. The experimental results show that the MTM is more effective for the task of SMS spam filtering.