Abstract: The current cooperative spectrum sensing based theory of evidence think all cognitive users are honest and doesn’t consider malicious users existing. When malicious user falsifies the local sensing result and sends error evidence to the data fusion center, it will reduce the cooperative spectrum sensing performance, which is called spectrum sensing data falsification(SSDF) attack. Due to the difference between the malicious user’s evidence and other cognitive users’, we use the Jousselme distance to calculate the credibility of evidence and proposes an SSDF attack detection algorithm. When the data fusion center receives all cognitive users’ evidence, it excludes the evidence with lower credibility and combines higher credible evidence weighted according to credibility. As simulation results shown, the SSDF attack detection algorithm can improve the performance of cooperative spectrum sensing with malicious users existing.