Because of the dynamic and open characteristics of the cloud computing, continuing malicious attacks happen frequently. Combining the idea of trusted cloud, a trust-based defensive system model for cloud computing has been constructed to guarantee the cloud security. Through real-timely monitoring, users’ behavior evidences have been obtained and standardized; a new method for users’ trust evaluation based on fuzzy AHP (Analytic Hierarchy Process, AHP) has been presented, it gradually determines the weights of behavior evidences, achieves quantitative assessment of behavioral trust; to provide great security defense for users, multiple detection engines have been used to conduct a comprehensive inspection of suspicious files and integrated decisions have been made. Experimental results show the system model can effectively eliminate the malicious behaviors from undesirable users, reduce users’ damages caused by virus and achieve a two-way defense for both cloud and client.