This paper presents a feasibility study on the use of multilayer neural networks to determine airflow infiltration from thermographs, thus the related energy leak. The developed method, aimed at accurate evaluation of intake airflow through an opening in the building's envelope, uses as input data infrared images of the temperature changes in a rendering surface near an opening in the building's envelope. Data collection of these measurements can be achieved with relative simplicity, and therefore could lead to an alternative or complementary method to the standardised ways of measuring infiltration based on Blower Door test, increasing the possibilities of monitoring, supervision and continuous management of building's ventilation and airtightness. Laboratory results show over 93% average accuracy for instant samples, and over 98% global accuracy for sequences. The generalization capability of this method has also been explored, and several neural network topologies analysed. (C) 2019 Elsevier B.V. All rights reserved.