Numerous authors have reported heat transfer prediction using artificial neural network (ANN). However, the precision or accuracy of the calculation is generally unknown. Error propagation from Monte Carlo method is applied to the coefficient of performance (COP) predicted by ANN. This COP permitted us to evaluate a water purification process integrated into a heat transformer. A feedforward network with a hidden layer was used in order to obtain error propagation in COP prediction. This model used the input and output-temperatures for each component (absorber, generator, evaporator, and condenser), as well as two pressure parameters from the absorption heat transformer and LiBr + H(2)O mixture with different LiBr concentrations. The hyperbolic tangent sigmoid transfer-function and the linear transfer-function were used for the network. A new correlation for calculating relative standard deviation (%RSD(COP)) of COP as a function of COP(EXP) and %RSD(instrument) was obtained. This study shows that %RSD(COP) of ANN prediction decreased when the experimental COP is increased. The range of COP operations was from 0.21 to 0.39. (C) 2010 Elsevier Ltd. All rights reserved.