Abstract: To resolve the problem that traditional primary user sensing technique cannot detect PUE (Primary User Emulation) attack in cognitive radio networks, a novel detecting technique is put forward that based on SVDD (Support Vector Date Description). Detecting PUE is modeled as a one-class imbalanced classified problem in this technique. Without PUE attack parameters, a one-class classifier is trained with legal users’ parameters by SVDD algorithm. Input the pending samples to the trained classifier, PUE will be detected. Analysis and simulation results show that, a lower false alarm probability and miss detection probability can be obtained.