Porous metal fiber sintered felt (PMFSF) is a promising critical component in proton exchange membrane fuel cells, having the ability of simultaneously acting as the flow field plate, gas diffusion layers and also the catalyst layers support, and owing the property of multi-scale surface morphology. A simple multi-scale mathematical method was proposed to actively construct three dimensional models of PMFSF's microstructure by synthesizing the implicit periodic surface (PS) model and the Weierstrass-Mandelbrot (W-M) fractal geometry. In this method, the PS model described the macro overall fiber shape, and the W-M fractal geometry modeled the micro fractal roughness topography attached. Based on the method, multi-scale fractal PMFSF models were reconstructed according to morphology parameters of physical PMFSFs, and were discretized in ANSYS/ICEM to generate refined mesh for computational fluid dynamics analysis. To verify the validity of the proposed modeling approach, PMFSFs with different porosity and fiber orientation are generated, and then the effects of the fractal surface topography and the fractal parameters such as fractal dimension and height scaling parameter on the gas permeability of PMFSF were investigated. The numeric simulation results show that the influence of the fractal topography on the in-plane and through-plane permeability of PMFSF cannot be ignored, and the permeability of fractal PMFSF models agrees with experimental measurements better. Especially, the results imply that the fractal morphology may have the potential to adjust the anisotropic properties of PMFSFs' permeability. It is further found that the larger the fractal dimension is and the lower the height scaling parameter is, the better the permeability of PMFSF will be. The synthesis approach and numerical simulation method may facilitate the development of active functional design mode to predict and optimize key characteristics of PMFSF ahead of manufacture. Copyright (C) 2016, Hydrogen Energy Publications, LLC. Published by Elsevier Ltd. All rights reserved.