The technology of spectrum sensing plays a key part in cognitive radio(CR). This paper proposed the SOMSVM spectrum sensing model. This model takes the advantage of the clustering of SOM which can contain the same input samples together and decrease the input samples near by the center of the other clustering. SVM is trained by the representative training set after compressed by 20% in SOM. In this paper, sample sets are generated by laboratory instruments, verifying that the SVMbased spectrum sensing model also can obtain good performance under the condition of actual communication environment. The simulation results show that in low SNR the probability of detection is 100% and the probability of error detection is improved by SOMSVM spectrum sensing model.