Learning control is a method of compensating for reproducible errors due to repetitive movements by feedforwarding. The greatest advantage of this control method is that even if the dynamics of the controlled object is unknown, compensation can be performed from the input / output relationship. However, the conventional learning control is a method for robot arms with little time delay and it was difficult for providing sufficient compensation of the trajectory error of the hydraulic excavator which includes large time delay in its control system. Therefore, we considered and proposed the effective learning control method which could be applied to a system with a large delay. It is averaging the output error from each control time to the system delay time ahead. In this paper, we verified the basic effectiveness of proposed learning control by applying it to the experimental set-up of a hydraulic excavator.