The signal sparsity is often unknown, or even changed with time in spec trum sensing. Therefore, a variable step size sequential compressed de tection algorithm is proposed by combining the adaptive theory with the sequenti al compre ssed spectrum sensing technology. The functional relationship is established bet ween step size factor of the next needed measurement numbers and current distan c e from the likelihood ratio and the detection thresholds. In addition, the sh ortcoming of the fixed step size of the measurement increment in the existing se quential compressed sensing is overcomed in the proposed algorithm. Theoretical analysis and computer simulations are conducted by introducing the rules of step size adjustment with piecewise function and parabolic function, respecti vely. Simulations prove that, the proposed algorithm has the fast er detection speed, less measurements number and lower computational complexi ties, compared with the existing sequential compressed detection scheme.