This paper treats the estimation of human impedance and their application to human-robot collaboration work. Initially, we perform an experiment at which the human becomes a slave and the robot behaves like a master having F/T sensor on its end. The human impedance expressed in terms of mass, damping, and stiffness properties are estimated based on the force data measured by F/T sensor and the commanded position data of the robot. To show the effectiveness of the estimated human impedance, we perform the second experiment at which the roles of the human and the robot are reversed. It is shown that the robot using the estimated human impedance follows the trajectory commanded by human very smoothly.