The optimization of a nonlinear adaptive multi-channel active noise control (ANIC) system in a rectangular enclosure using neural networks is investigated in this paper. The model of enclosure is obtained using modal analysis techniques, and the target bandwidth of the control system for global reduction of noise is selected to be 50-300Hz. Secondary path in modeled offline using multilayer perceptron (MLP), and standard back-propagation algorithm by choosing a multi-tone as an excitation signal. The simulation results assuming linear and nonlinear models of secondary path show that in single-channel case multilayer perceptron neural networks with FxBP algorithm have superior performance than FIR structure with FxLMS algorithm, and in multi-channel case their performance are comparable.