Abstract: The problem of line spectrum detection in shallow water is investigated in this paper.The statistics of ambient noise recorded in sea trials is firstly analyzed by distribution fitting and Kolmogorov-Smirnov test,and the generalized Gaussian distribution is proved to be an appropriate model for ambient noise modeling.Thereafter,the generalized maximum likelihood detector(MLD) and asymptotic detector are deduced and the maximum likelihood estimation of frequency is proposed under the assumption of weak signal.The performances of three types of line spectrum detectors have been verified by simulation,and the results show that the accuracy of noise model plays an important role for signal detection and the MLD is an optimal one for line spectrum detection in ambient noise.