Abstract: By designing the input signals of the system, the nonlinear part of the Hammerstein system can be removed from the system when the linear part of the system is being identified, the linear dynamic part can be obtained. Using the obtained linear mode, a filter can be constructed to reconstruct the intermediate inputs according to the outputs. So the nonlinear part can be estimated by some smooth technologies. A multivariable Hammerstein system identification is proposed based on the dynamical separation method. The simulation results show its effectiveness.