In this paper, a biped robot system with the unmodeled dynamics is approximated by the weighting combination of N fuzzy-based linear state-space subsystems which are achieved from the linearization of the biped robot system about a desired trajectory. Then, the same fuzzy sets of the system rule are applied to establish the fuzzy reference models with desired amplitude and phase properties. The reason to introduce the reference model is not only to obtain a tracking error system, but also to guide the suitable response for the dynamic walking of the biped robot. The system uncertainties contain: the approximation error of the fuzzy-model, the unmodeled dynamics of the biped robot, the discontinuity of velocity as contacted with ground, and the interaction dynamics resulting from the other subsystems. Based on the concept of parallel distribution compensation, a fuzzy-model-based sliding-mode control is designed. The proposed control includes a fuzzy equivalent control and a fuzzy switching control. The stability of the overall system is then verified by the Lyapunov stability theory. The illustrative example of three-link biped robot is given to explain the design procedure.