The invention relates to a spectrum sensing method. In this method, the spectrum sensing problem is modeled as a signal classification problem. After data processing and feature extraction, the support vector machine is used as a classifier to realize spectrum sensing. The steps are as follows: firstly, collect signals in the presence and absence of primary user signals respectively; secondly, extract features and mark labels; thirdly, optimize parameters and use training data for learning to obtain the optimal classification hyperplane; Fourth, extract the characteristics of the signal to be tested, and use the obtained classification hyperplane for discrimination to achieve spectrum sensing capabilities. For the convenience of description, the ratio of the maximum and minimum eigenvalues of the signal covariance matrix is selected as the classification feature. In specific applications, other features, such as signal energy, signal spectrum, and cyclic spectrum, can also be selected as classification features. The invention not only solves the problem that the decision threshold is difficult to set in the common spectrum sensing method, but also has superior sensing performance.