The process of heat transfer within porous media is Usually considered as a transport through large numbers of straight channels with uniform pore sizes. For the prediction of effective thermal conductivity of gas diffusion layer (GDL), morphological properties such as the tortuosity of channels and pore-size distribution of this porous layer should be considered. Thus in this article, novel parallel and series-parallel prediction models of effective thermal conductivity for the GDL in proton exchange membrane fuel cell (PEMFC) have been derived by fractal theoretical characterization of the real microstructure of GDL. The prediction of fractal parallel model for carbon paper, a basal material of the GDL, is in good agreement with the reference value supplied by Toray Inc. The prediction results from the proposed models are also reasonable because they are distributed between the upper and lower bounds. Parametric effect has been investigated by using the presented models in dimensionless formalism. It can be concluded that dimensionless effective thermal conductivity (k'(eff)) has a positive correlation with effective porosity (epsilon) or the pore-area fractal dimension (D-p) when k(s)/k(g) < 1; whereas it has a negative correlation with epsilon or D-p when k(s)/k(g) < 1 and with tortuous fractal dimension (D-t) whether k(s)/k(g) < 1 or not. Furthermore, these fractal models have been modified by considering the effect of polytetrafluoroethylene (PTFE) incorporated into the pore spaces of carbon paper, and the corresponding model prediction shows that there is an increase in the effective thermal conductivity due to the filling of PTFE that has high thermal conductivity. (c) 2008 Elsevier B.V. All rights reserved.