Neural network models as well as fuzzy systems have been shown capable of approximating functions and thus for identification of complex systems. In the present paper, a new neuro-fuzzy architecture, called FasArt (Fuzzy Adaptive System ART based) is proposed. It enhances and adapts the Fuzzy ARTMAP neural-fuzzy architecture for nonlinear systems identification, that was initially proposed for supervised pattern recognition. It emphasizes the complementary nature of neural networks and fuzzy systems, while it provides design parameters in order to control the fuzziness degree of the fuzzy sets and the distributed character of the information. The architecture with proven stability and capacity of approximating a function, was tested experimentally with satisfactory results in the highly complex problem of penicillin production and in approximating a function, typically used for comparison in the literature.