A new nonparametric method to identify multivariable Hammerstein models is presented. The Hammerstein model is characterized by a combination of a linear dynamic subsystem and an algebraic nonlinear function. There could be many different models that give the same input-output realization. The purpose of this identification is to find out one among those models for controller design. This identification uses a sequence of specially designed test signals for excitation. The linear dynamic subsystem is identified as a finite sequence of impulse response (FIR), and the static nonlinearity is identified as a mufti-input-mufti-output (MIMO) functional mapping. By making use of this special test signal, the FIR sequence can be estimated under a single-input-single-output (SISO) framework. Moreover, the identification for linear subsystem can be decoupled from that for the nonlinear static part. This nonparametric model can be used for model predictive control applications.