Composite electrodes for solid oxide fuel cells were modeled by three-dimensional resistor networks. The networks were generated on a computer by identifying neighbors in either cubic lattices randomly occupied by electrolyte or electrode particles, or in random packings generated by sequential deposition of such particles in random order. The resistance between any two particle sites i and j were taken to be functions of the conductivities of the particles residing in site i and j and the periphery of the necks formed between them. An emphasis was put on parameters believed to be relevant for cermets of Ni and yttria-stabilized zirconia. The conductivity df the networks were calculated numerically, and the results of the model are in good agreement with experimental findings. A sharp transition from low to high conductivity occurs at approximately 30 volume percent of electrode material for cubic lattices and for the random packings with uniform particle radii. In the bimodal random packings, this percolation threshold increases with increasing electrode-particle radius relative to the electrolyte-particle radius. This is suggested as a possible explanation for cermet deactivation under operation, since upon aggregation of electrode particles the percolation threshold may increase past the given volume fraction of electrode material in the composite.