Abstract: The prediction and filtering of underwater acoustic signals, especially of reverberation interference and back-ground noise, lay the foundation of underwater target signal detection and processing of non-stationary, non-Gaussian and nonlinear underwater acoustic signals. In this paper, based on the linear theory of least square estimation and the Volterra series theory, two prediction models of target signal are established to conduct one-step and multi-step predic-tions. Thus, the optimal prediction parameters are obtained by comparing and analyzing the prediction results. Simula-tions show that the prediction model based on singular value decomposition of Volterra series achieves more accurate results in predicting the underwater acoustic signal, and its relative error of prediction is one-order of magnitude smaller, compared with the model based on least square estimation.