Abstract: In underwater acoustic communication, intersymbol interference caused by multipath effects can be mitigated using equalization. Blind equalization without training sequence is a bandwidth efficient way to solve this problem. In fact, communication channels are not completely linear. Neural network is a nonlinear dynamic system, which can be realized with large-scale parallel processing and has good robustness. Forward feedback neural network(FNN) using a variable step size BP algorithm is used to implement underwater acoustic channel equalization. Results of computer simulation indicate that, by using the proposed algorithm, higher convergence rate and better performance are obtained compared to the traditional BP algorithm.