If occupants of buildings are offered possibilities to interact with the building's equipment elements such as with windows - in order to optimize their individual environment, these interactions will influence the energy consumption of the building. Therefore, during the design of the building, e.g. by building simulations, these interactions need to be predicted if the energy consumption of the building is to be optimized. These interactions are partly motivated by the need for thermal comfort. A precondition for the prediction of interaction is therefore the prediction of the individual evaluation of the thermal environment. Although 'sensation' is not an optimal conceptualization of 'satisfaction with the thermal environment', it is frequently used as a measure for the evaluation of thermal comfort. However, the prediction of thermal sensation is currently not satisfactorily possible. Therefore, this article examines the potential of artificial neural networks to improve the predictability of thermal sensation. The data base used for this research derives from the RP-884 Adaptive Model Project. Results show that the designed neural network performs excellently in the prediction of the distribution of individual ASHRAE votes under defined conditions, and that it outperforms the classical PMV index in terms of prediction quality and the range of information contained in the prediction. (C) 2015 Elsevier Ltd. All rights reserved.