The identification of nonlinear MIMO systems of the Hammerstein kind is discussed in the paper. The procedure is based on the approximation of the linear system dynamics by Laguerre filter banks, and on the approximation of the static nonlinearities by neural networks. The parameters of the approximating structure are identified from input-output noisy data with the modulating functions method. Under some restrictions on the MIMO structure of the system the proposed identification procedure is shown to be convergent and robust to measurement noise.