Spectrum sensing is the basis of cognitive radio technology. Cooperative spectrum sensing has been shown to increase the reliability of spectrum sensing. To reduce sensing overhead and total energy consumption, it is recommended that the users with good performance should be selected to increase the sensing reliability. However, which of the cognitive users have the best detection performance is not known a priori. In this paper, a selective cooperative sensing strategy and a user selection method based on B value are proposed so as to increase the sensing reliability and reduce sensing overhead. Simulations are used to evaluate and compare the B value method with other methods. Simulation results show that B value selection has the same sensing performance with signal-to-noise ratio (SNR) selection. B value selection obviously outperforms the simple counting selection in the presence of noise uncertainty. In general, the SNRs of all cognitive users are not known a priori and there must be certain noise uncertainty, in this sense, B selection is a simple, feasible and effective selection method.