In this study, we explain the causes and effects of the geometrical impossible result encountered in the widely adopted tetrakaidecahedron model (Boomsma and Poulikakos, 2001; Dai et al., 2010) for the effective thermal conductivities (ETCs) of metal foam. The geometrical impossible result is successfully eliminated by accounting for the size variation of the node with porosity. The improved model provides predictions of ETCs that are more precise than available models. For aluminum foams (ks=218Wm-1K-1) using water and air as fluid media, the relative root-mean-square (RMS) deviation of the present predictions from the experimental data is about 5.3%; for the reticulated vitreous carbon (RVC) foams (ks=8.5Wm-1K-1), the relative RMS deviation is about 7.4%.