A compressive autocorrelation detection al gorithm is proposed for overcoming the detection problem of unknown high bandwid th sparse signals. Firstly, the compressive sensing technology is unutilized to acquire the signals at a sampling rate which is far lower than Nyquist samplin g rate. Then based on researching autocorrelation matrix theories of signal dete ction, a sparse coefficients compressive autocorrelation detection algorithm usi ng statistical distribution is deduced through the restricted is ometry property of the sensing matrix and the compressive samplings are dealt wi th d irectly. The connection is subsequently obtained between the decision threshold and the f alse a larm probability theoretically. Moreover, the computational complexity of the algori thm is analyzed. Therefore, the method can improve the dete ction timeliness efficiently through few compressive samplings without reconstruct ing signal. Simulations show t hat the proposed algorithm still perform well in unknown signal detection with l ow signal to noise ratio.