In this paper, a human arm-like robot control scheme is proposed based on time-varying viscoelastic properties which consist of multi-joint stiffness and multi-joint viscosity during human arm movements and a modified forward gaze model. In general, in human multi-joint arm movements, the multi-joint torque is assumed to be a function of multi-joint stiffness matrix, multi-joint viscosity matrix, and motor command descending from central nervous system (CNS). In order to make the present human arm-like robot move like a human multi-joint arm, a feedback controller and a modified forward gaze model are presented in the human arm-like robot control system. That is, the feedback controller is designed to obtain desired motion mechanism based on real measured data from viscoelastic properties of human multi-joint arm, and the forward gaze model in which steering gains are modified using a cost function is used to compensate the term related to the effect of CNS. The effectiveness of the proposed method is confirmed by the simulation results based on experimental data.