Abstract: Due to serious attenuation and distortion of acoustic signal caused by multipath effect and frequency dispersion effect in underwater acoustic channels, the traditional equalization techniques cannot meet the application requirements in underwater acoustic channel. Outstanding performance of neural network in equalization techniques has attracted widespread attention in recent year, so an efficient neural network training algorithm, known as the improved symbiotic organisms search algorithm based on nonlinear autoregressive neural network with exogenous inputs (NARX-nSOS), is proposed for underwater acoustic channel equalization in this paper. The algorithm is optimized with the symbiotic organisms search (SOS) algorithm based on nonlinear autoregressive neural network with exogenous inputs (NARX) equalizer and its convergence capability is improved by combining opposition-based learning (OBL) method. The effectiveness of the NARX-nSOS is verified by computer simulation, and the results demonstrate that the NARX-nSOS algorithm can accelerate convergence speed and improve communication quality significantly.