This paper presents a multiple-model hypothesis testing approach based on sequential probability ratio test for cognitive radio spectrum sensing to detect unknown signal that may have multiple possible distributions with different structural or parametric uncertainties. The traditional cognitive radio spectrum sensing scheme (e.g., single model hypothesis testing based on sequential probability ratio test and M-ary hypothesis testing) may be not correct, because it only handles the totally known signal distribution case without considering the uncertainties of signals. The proposed multiple-model hypothesis testing scheme not only copes with the uncertainties of signals, but also has a setting that can provide efficient detection results. Performance of the proposed scheme is evaluated for spectrum sensing in an illustrative scenario. Simulation results demonstrate its detection efficiency compared with the traditional schemes.