Based on t-distribution and Anderson-Darling test,the first,a method was proposed of spectrum sensing under assumptions that the noise variance was unknown and also the channel information was unknown.Indeed,the sensing was conducted based on Anderson-Darling distance between the distribution of ratio of sample’s mean to sample’s variance and the t-distribution.The second,another sensing method was proposed,under the same assumptions as the above,based on Anderson-Darling distance between the empirical characteristic function and a known characteristic function.The third,lower bounds on detection probabilities of sensings above were given for a flat,slow fading channel.Theoretical analysis and simulations show that the performances of proposed methods are much better than energy detector-based sensing with known noise variance,especially under an environment of a low signal to noise ration(SNR),or small number of samples.