Aiming at the characteristics of complex, time-delay and gain time-variation of agricultural robot steering system, the Adaline neural network (ANN) was applied to internal model control (IMC). An adaptive- control method for time-delay online adjusting and gain control was proposed. An algorithm based on Adaline neural networks could adjust time-delay and gain adaptively, and overcome effects of time-variation parameters on IMC and plant model by online optimizing feedback error. Simulations and experimental results verified that compared to conventional PID control method, the proposed control method possessed the advantage of high precision, great adaptability and robustness, so it is feasible for agricultural robot steering system.