Many exoskeletons for lower limb assistance are actuated by electrical motors. Also, passive elements are often used additionally to provide extensive negative power of the exoskeletal joints and reduce the burden of the actuators. However, the passive elements sometimes work in resistive manner, rather than assisting the motion of wearer due to those poor controllability. Therefore, in this paper, passive element (i.e., spring) of the exoskeletal knee joint are opti- mized for the walking with different loads. Through the dynamic analysis which consider the interaction between models of human and exoskeleton, the optimal spring constant was determined such that the joint torques of human model are minimized. For the dynamic simulation, MATLAB SimMechanics was used. The walking motion with three different load conditions (0, 25, and 40 kg) were obtained by three-dimensional motion analysis and were used as motion inputs for inverse dynamic analysis. The simulation results showed that human joint torques were reduced by 16% with the optimal spring stiffness. For each load condition, however, different optimal spring constants were observed. This imply that the realization of variable stiffness is necessary for design of exoskeleton as a future work.