Abstract: Spectrum sensing is the key technology in cognitive radio field. This technology is enable the secondary users to detect the underutilized spectrums "white space" which have not been occupied by the primary users, improving the spectrum efficiency of the whole system. Wideband spectrum sensing requires several GHz bandwidth sensing. Excessively high sampling frequency and large amount of data are the major challenge for existing hardware devices. By utilizing the sparsity of wideband spectrum, this paper proposes a new spectrum sensing method based on OMP algorithm for wideband spectrum sensing. In the proposed method, MWC sampling is used to implement compress sampling for the wideband analog signal directly. The compression sample model with finite dimension is obtained by using the symmetry decomposition property of autocorrelation matrix and the independence of the primary user’s signal. Besides, AIC/MDL criteria is used to estimate the sparsity which is a threshold of the stop iteration for the OMP algorithm. As a result, the complexity of the algorithm is reduce greatly. The estimation of the signal’s PSD is skipped in our method. The occupied channels are detected directly from the compress sampled data in time domain at low rates. Simulation results show that when the in-band SNR is above 9dB, the spectrum detection probability is greater than 90%.