The correct estimation of the energy performance of windows represents a fundamental issue in the buildings heat losses assessment, since both accuracy and simplicity requirements have to be fulfilled. The most common approaches proposed consist of finite element simulations and laboratory measurements. The reliability of numerical methods results is still an open issue: the general policy of Standards requires for the calculated values to be on the safer side but it is not clear if this assumption is always verified and which is the difference between measured and simulated values. Standards allow the implementation of numerical simulations also in a two-dimensional domain, while windows are composed by zones with a definite three-dimensional geometry such as corners. Two types of windows (wood and aluminum framed) were analyzed, showing that the difference between two-dimensional and three-dimensional simulations is not so high to justify the vast difference that exists instead in terms of calculation time. By means of a hot box apparatus, a comparison term was given to evaluate the finite volumes method adherence to the experimental values. Infrared thermography gives the possibility of a better understanding of the differences between the simulation results and the measured values. The experimental measurements indicated a good data coincidence with simulations for both types of Windows, with an expected thermal window performance underestimation of the numerical approach. The reasons lie in the use of thermal conductivity cautionary values in simulations, and on the stratification phenomenon (highlighted by the infrared camera), that brings a decrease to the convection heat transfer. It emerges therefore that, if a window manufacturer strives for achieving high performance certified products, the test on an experimental hot box setup becomes more appropriate, since it brings to a thermal resistance evaluation around 10% higher than the one obtained from a simulation procedure. (C) 2014 Elsevier Ltd. All rights reserved.