Effective thermal conductivity in 3D heterogeneous media consisting of fillers dispersed randomly in a matrix is evaluated through FEM based numerical simulations. The study examines the influences of various factors such as the filler concentration (up to 30%), fillers shape (spheres and polyhedrons) and size (100 to 300 μm) and thermal characteristics of fillers and matrix. Simulations indicate that (a) particle distribution plays a significant role for moderate filler concentrations, (b) non-spherical particle shape has a strong influence in the heat conduction, (c) higher the ratio of individual thermal conductivity of fillers to matrix, greater is the thermal conductivity and higher is the sensitivity to the distribution. The numerical assessment is compared with predictions from effective medium theories and published experimental data for Silica (SiO2)/Epoxy and Alumina (Al2O3)/Epoxy polymer composites. The model was found to be robust to capture experimental data. © 2020 Elsevier Masson SAS