The loss of fine particles can induce mechanical instabilities in granular soils subjected to internal fluid flow. An appealing countermeasure consists of the re-injection of fine grains with the objective of achieving retention in the soil matrix. In this study, both gravity- and fluid-driven infiltration of fine particles into coarse-grain columns with different solid fraction ϕ and size ratios R have been studied using coupled Pore-scale Finite Volume (PFV) and Discrete Element Method (DEM) schemes. Three clogging regimes, surface clogging, deep filtration, and percolation are detected, and the characteristic infiltration depths L0 are found to grow exponentially with R under gravity- and fluid-driven cases. A probabilistic model derived from pore-constriction size statistics is then put forward, which could efficiently interpret the decaying distribution of fine retention for a given size ratio R and packing density. The mean transit velocity of fine grains follows an increasing trend with √ R under fixed ϕ and can be collapsed over an almost constant value with the appropriate scaling of ϕ/ R. Compared to gravitational percolation, more lateral dispersion is found in fluid-driven conditions, and an estimation of the related lateral dispersion coefficient D is provided based on ϕ and R.