This paper presents a variable structure based approach to the load frequency control system problem in electric power generation systems. This approach combines the salient features of both variable structure and fuzzy systems to achieve high-performance and robustness. The control signal consists of equivalent control, switching control and fuzzy control components. The influence of system nonlinearities such as generation rate constraint and governor dead band is considered. In addition, a decentralized form of this controller is applied to a two-area power system. Simulation results show that the system responses are strongly robust for parameter variations.