The parallel robot, also known as parallel kinematic machine (PKM), has some unique properties such as higher speed, stiffness, and load carrying capacity compared with serial robots. However, the dynamics of the PKM is normally more complex than that of serial robots, due to the highly coupling relation between the moving components. Hence this property affects the control performance of the PKM. To address this issue, the dynamic identification of the PKM is discussed for dynamic visual servoing in this paper. A visual sensor-based dynamic identification method for a 6-DOF revolute-sphere-sphere (6-RSS) PKM is proposed. In contrast to the classic method for PKMs, the proposed method doesn't require the actuator torque measurement and the parameters of the internal robot controller. The forward kinematics of PKMs are not needed. In this paper, the principle of virtual work method is utilized to build the dynamic model of the 6-RSS PKM. A visual sensor based closed-loop identification algorithm is developed to estimate the dynamic parameters of the PKM. The experiment tests show that the output of the identified model matches that of the PKM testbed with satisfactory accuracy when both systems are subjected to the same desired path.