A control-oriented multivariable Hammerstein model is used to identify the strongly nonlinear dynamics of fuel cell systems that are described by nonlinear differential or differential-algebraic equations. Within the Hammerstein model framework, the static nonlinear part is constructed by a wavelet network, and the linear dynamic part is described by a discrete-time transfer function of the state-space model. For prescribed input-output patterns for high-order fuel cell systems, simulations demonstrate the accuracy of system identification using wavelet networks in the Hammerstein structure that is better than that in the neural network structure.