A kind of fast linear generalized predictive control (GPC) algorithm is proposed based on the extended state observer for chaotic (hyperchaotic) systems. The linear extended state observer is employed to estimate and compensate the nonlinear dynamics and the existing uncertainties of the chaotic (hyperchaotic) systems so that an integrator can be obtained to serve as the model for GPC design. Using this scheme, the computational complexity can be substantially reduced. A step coe?cient matrix can be derived analytically and a future output prediction can be explicitly calculated by only using the last two samples of the output. Therefore, the self-tuning algorithms and the Diophantine equation can be completely avoided. The proposed method can be used to control nonlinear targets in a straightforward manner. Simulation results show the effectiveness of this linear algorithm.