To design a controller for the purpose of control of hand exoskeleton force, the critical question is to model the human finger impedance properly. The difficulty is that the parameters of the impedance model are perturbative in different postures. In this paper, multiplicative uncertain models of human finger impedance are presented. We describe the impedance as a mass-spring-damping system. The experiments are set in extension, half-flexion, flexion and synthesis postures. The parameters of nominal model as well as their perturbation range are identified by the forgetting factor recursive least square method. The weighting functions are constructed accordingly. The results show that the uncertain model of synthesis posture can represent the magnitude-frequency characteristic at low frequency, while, for high frequency, the uncertain models composed by the weighting function of synthesis posture and the corresponding nominal model are feasible. These provide effective approaches for hand exoskeleton force control issues.