Accurate knowledge of the thermal properties of building components is necessary to implement adequate energy saving strategies in buildings. Outdoor experiments using test cells are very useful tools for realistic estimation of these properties. This paper describes the analyses performed, and the procedure followed in identifying and solving some problems found when building components are tested for UA and gA in a test cell under warm and moderate weather conditions. A window component was tested in a PASLINK test cell at the CIEMAT's 'Plataforma Solar de Almeria (PSA-CIEMAT)' in Tabernas (Almeria, Spain) and several data sets recorded under quite different weather and test conditions were analysed. First the problems identified when applying the usual test and linear analysis procedures are described. Then hypotheses about the cause of these problems are formulated. Afterwards, strategies followed for testing these hypotheses are described. Once the cause of the problems had been identified, they were fine tuned to find a model for accurate UA and gA estimation. This study demonstrated that nonlinear models, in which long wave radiation is considered as nonlinear effect, yield remarkably better performance than the commonly used linear models, for estimating the component UA and gA values. (C) 2008 Elsevier Ltd. All rights reserved.