To resolve the difficult problems in robot fish dynamics modeling such as transient strongly nonlinear flow control, establish the robot fish straight swim steady-state velocity model based on general regression neural network (GRNN). Taking the three joints biomimetic robot fish as the researching object, use the nonlinear approximation capability of the neural networks, make use of GRNN to recognize the strongly nonlinear relationship between robot fish swim velocity and its motional parameters, set up the relation model of the motor controlled parameters and the biomimetic robotic fish straight swim steady-state velocity, carry out experiments to do the error analysis between the predicted and actual values. The experimental results prove that the biomimetic carangiform robot fish straight swim velocity model using the GRNN neural network identification technique is feasible.