In this paper, efforts are made to improve upon the existing structure of fuzzy neural network (FNN) in order to enhance its whole characteristics and then the heat exchange performance of certain air conditioning unit was modeled by the advanced FNN. Not only was the heat transfer rate interpolated when the chilled water is at 6 and 8℃ separately on the basis of that available when chilled water is at 5、7 and 9℃, but also the heat transfer rate is extrapolated in the state of the air flow rate reaching 1.33m3/s in according to the data known while the air flow rate is 1.0m3/s and 1.25m3/s. All relative errors are less than 5% when the modeling output is compared with the real. Analyses show that the more accuracy with faster convergence speed and the better generalization are obtained if the fuzzy technique is applied to the neural network instead of using artificial neural network (ANN) only.