This paper presents an asymmetric stiffness characteristic of a human hand. In human support robotics or medical robotics, the detail comprehension of physical human body is important to develop safe and high performed robots to work cooperatively with a human and to replace human dexterous tasks. It is known that a human arm generates variable stiffness depending on tasks by coactivation of agonist and antagonist muscles. Previous related researches have been presented impedance characteristics of a human upper limb in static posture and dynamic motion. These characteristics are represented by ellipsoids. However, the above analyses are based on a simple muscle model and conventional kinematic and dynamics of an articulated body system. In this paper, perturbation-excited method is carried out for estimating the stiffness of a human hand. The experimental results demonstrate nonlinear property of the stiffness of a human hand. To illustrate the observed stiffness characteristic, this paper proposes nonlinear stiffness model of the human hand.