The dynamic behavior of granular filtration is investigated in the present paper by using the Brownian dynamic simulation method to track the individual particles moving through the 2-D network model of porous media in the filter bed. The scope includes three different network structures, modified square, triangular and honeycomb, all with the same initial permeability, and two types of DLVO interaction energy curves. The specific filtration functions of F-1 (alpha', sigma) and F-2 (beta', sigma) are determined from the breakthrough curve of effluent concentration and the pressure drop. When comparing those three different network models with the same interaction energy curve, it is found that the honeycomb network can filtrate particles the easiest, because of its lowest coordination number, so its corresponded specific deposit a is the highest. When comparing with the available experimental data under the unfavorable deposition conditions, it is found that the present triangular network gives smaller discrepancy than that of the convective diffusion model. (c) 2006 Elsevier B.V. All rights reserved.