With the goal of constraining fluid and melt transport rates in the upper mantle and lower crust, various permeability--porosity relations have been proposed for deep-seated, microstructurally equilibrated rocks. Two relations that are in close agreement include one that is based on numerical modeling for an idealized rock and another that is based on direct measurement of permeability on synthetic quartzite analogs. Each relation describes a continuous increase in permeability as a function of porosity, raised to some power. The empirical relation is displaced to lower permeabilities, reflecting the more complex (nonuniform) pore geometry of the quartzites, which resembles that expected in natural, deep-seated rocks. Despite differences, the numerical models assume, and the quartzites display, a similar pore geometry that is consistent with thermodynamic equilibria: grain boundaries are dry, with pores largely confined to joins of three or more grains. In contrast, a fundamentally different pore microstructure, in which most of the melt occupies grain boundaries, has been proposed for partially molten dunite. This has led to the suggestion of a unique, noncontinuous, or threshold permeability relation for upper mantle melts. A more critical analysis of pore microstructure in dunite indicates, however, that pore shapes had been misinterpreted: very little melt is present along grain boundaries. Melt instead occupies a triple-junction network closely resembling that of the quartzite analog. Consequently, a relation similar to that for synthetic quartzites can describe upper mantle grain-scale permeabilities, because permeability is a function of pore geometry alone.