This article investigates the identification of interconnected block models with hard input nonlinearities. The cascated static nonlinear function followed by a linear dynamic representation is named Hammerstein model. The static nonlinearity is portrayed by a neural network that is simple and has accurate tuning capability, and the dynamic block, is represented by a state-space model that simplifies the extension to the multivariable case. Taking these characteristics into account, an approach was developed to identify a Hammerstein multivariable Neuro-Fuzzy model through a noniterative procedure associated with subspace identification methods. The functionality of the proposal was verified by simulation, yielding improved performance compared to the case of polynomial static nonlinear curve.