The objective of this study was to create a microfluidic model of complex porous media for studying single and multiphase flows. Most experimental porous media models consist of periodic geometries that lend themselves to comparison with well-developed theoretical predictions. However, many real porous media such as geological formations and biological tissues contain a degree of randomness and complexity at certain length scales that is not adequately represented in periodic geometries. To design an experimental tool to study these complex geometries, we created microfluidic models of random homogeneous and heterogeneous networks based on Voronoi tessellations. These networks consisted of approximately 600 grains separated by a highly connected network of channels with an overall porosity of 0.11–0.20. We found that introducing heterogeneities in the form of large cavities within the network changed the permeability in a way that cannot be predicted by the classical porosity-permeability relationship known as the Kozeny equation. The values of permeability found in experiments were in excellent agreement with those calculated from three-dimensional lattice Boltzmann simulations. In two-phase flow experiments of oil displacement with water we found that the wettability of channel walls determined the pattern of water invasion, while the network topology determined the residual oil saturation. The presence of cavities increased the microscopic sweeping efficiency in water–oil displacement. These results suggest that complex network topologies lead to fluid flow behavior that is difficult to predict based solely on porosity. The novelty of this approach is a unique geometry generation algorithm coupled with microfabrication techniques to produce pore scale models of stochastic homogeneous and heterogeneous porous media. The ability to perform and visualize multiphase flow experiments within these geometries will be useful in measuring the mechanism(s) of displacement within micro- and nanoscale pores.