The paper presents a sensitivity analysis of a method for predicting the thermal conductivity tensor of highly heterogeneous building composites containing bio-additives. The proposed approach is based on solving the heat conduction problem at the microscale while taking into account the actual material morphology.  The microstructural data of the composites are obtained using micro-computed tomography ( \(\upmu \) CT) images. The method was tested using real samples of wood fibers and a cement binder composite commonly used in the production of wood wool cement boards (WWCBs). This material is characterized by wood fibers oriented in specific directions (i.e., along the width and length of the board), which in practical applications are typically perpendicular to the heat transfer direction (i.e., along the thickness of the board). Consequently, the material exhibits significant anisotropy in thermal conductivity. The numerical tool developed incorporates a method for processing \(\upmu \) CT data, including thresholding, selecting a representative elementary volume (REV) size, determining the physical and thermal properties of the composite components and tuning the model using experimental data. However, the selection of threshold levels limits to distinguish between different composite constituents is inherently arbitrary and is based on measurement data. Therefore, it is essential to assess the extent to which variations in these parameters affect the predicted effective properties of the composite in question. Understanding this is crucial for evaluating the reliability and accuracy of the proposed method.