A direct-fired absorption chiller systes more heat nenrgy than a vapour-comlpression chiller system of slmilar capacity. The combined tise of fuel energy and electricity consumption an important criterion is its performance evaluation. Many current research efforts on absorption chiller system are related to the search of an optimal supervisory control stategy to minimize the energy costs System simulation has been a contventional approach for the investigation. However. the task of developing an accurater system model through the component-modelling approach can be tedious Oversimplification in the process, and the nonlinear structure of the equation set as result. are often the limitations for raching reliable conveging solutions. The system-based artificial neural network (ANN) modelling approach appears to be an atternative. This article describes the process of deriving an ANN model of a commercial direct-fired double-effect absorption chiller system. The techniques and the ways to overcome the diffculties ln the training orocess are discussed.