Nonlinear neuro-models for a discharge air temperature (DAT) system are developed. Experimental data gathered in a heating ventilating and air conditioning (HVAC) test facility is used to develop multi-input multi-output (MIMO) and single-input single-output (SISO) neruo-models. Several different network architectures were explored to build the models. Results show that a three layer second order neural network structure is necessary to achieve good accuracy of the predictions. Results from the developed models are compared, and some observations on sensitivity and standard deviation errors are presented. (C) 2003 Elsevier Ltd. All rights reserved.