Abstract: A method is presented to train radial basis function(RBF) neural network with genetic algorithms. This method is used to calcalate the displacement and main dimensions of submarines. The results show the advantages of this fast training process, good comprehensive searching ability and high precission.