With the revolution of mobile devices and their applications, significant improvements have been witnessed over years to support new features in addition to normal phone communication including web browsing, social networking and entertainment, mobile payment, medical and personal records, e-learning, and rich connectivity to multiple networks. As mobile devices continue to evolve, the volume of hacking activities targeting them also increases drastically. Receiving short message spam is one of the common vectors for security breaches. Besides wasting resources and being annoying to end-users, it can be used for phishing attacks and as a vehicle for other malware types such as worms, backdoors, and key loggers. The next generation of mobile technologies has more emphasis on security-related issues to protect confidentiality, integrity and availability. This paper explores a number of content-based feature sets to enhance the mobile phone text messaging services in filtering unwanted messages (a.k.a. spam). Moreover, it develops a more effective spam filtering model using a combination of most relevant features and by fusing decisions of two machine learning algorithms with the Dendritic Cell Algorithm (DCA). The performance has been evaluated empirically on two SMS spam datasets. The results showed that significant improvements can be achieved in the overall accuracy, recall and precision of spam and legitimate messages due to the application of the proposed DCA-based model.