The field of unsaturated soil mechanics has recently seen the introduction of pore network models, which attempt to replicate the void structure of porous materials. They are robust physically-based simulation tools and have been used to simulate constitutive relationships like soil water retention curves (SWRC) and unsaturated hydraulic conductivity functions. This work aims to present a pore networkmodeling approach for predicting hysteretic SWRC at various stress states using only grain size distribution and porosity data of the granular soils. The soil sample subjected to given stress conditions is simulated using the Discrete Element Method to obtain a stable packing of spherical particles representing the soil structure. From this packing, an algorithm based on the medial axis is availed to extract a network of pores and throats that describes the geometry and topology of the void structure of the granular soils. Various pore-scale mechanisms like the piston-like advance, corner flow, pore body filling, and snap-off are then used to model fluid displacements at the pore scale to simulate SWRC along the drying and wetting paths. The modeled SWRC under various stress conditions is compared with the measured curves obtained for granular soils from the literature, and the predictions are in good agreement with the experimental results.