The advent of the super aged society and decline in earning power makes it critical to develop technologies that can support the limitations of our human abilities. We have developed Robot Suit HAL to support human physical capabilities. A wearable robot expected to work in human society is required to have a high structural safety, with a tough and compact structure and a human outline. To achieve this, this paper focuses on the control method for a wearable robot based on the structural safety. We suggest that HAL autonomously leads a wearer to a known appropriate working posture to reduce the structural stress based on prior FEM analysis. This is called Working Posture Control. The purpose of this paper is to propose a working posture control method and to verify the effectiveness of the proposed method. The target motion is the holding of a load with both forearms, such as is done during heavy work. The heavy work assistance means a higher workload for HAL than ever before, compared to usages such as the walking assistance. FEM analysis shows the waist component becomes stressed especially. Therefore, the reference working posture is defined as one where the upper body is inclined backwards using the principle of reducing the structural stress by shortening the moment arm. To this end, the control algorithm switches from control by bio-electrical signals to proportional and derivative control based on the angles and the angular velocities of the joints at the right time. When this method was applied in an experiment, the angles of the hip and knee joints follow the reference angles for changing to the appropriate working posture of the load holding motion and it was shown that the structural stress of the waist component was reduced. This confirmed the effectiveness of the proposed method.