In an attempt to obtain a conservative estimate of virus removal during slow sand and river bank filtration, a somatic phage was isolated with slow decay and poor adsorption to coarse sand. We continuously fed a phage suspension to a 7-m infiltration path and measured the phage removal. In a second set of experiments, we fed the phage suspension to 1-m long columns run at different pore water velocities. Using the data obtained, a mathematical model was constructed describing removal vs. pore water velocity (PWV), assuming different statistical distributions of the adsorption coefficient λ. The bimodal distribution best fit the results for PWVs higher than 1 m/d. It predicted a removal of approximately 4 log10 after 50 days infiltration at 1 m/d. At PWVs below 1 m/d the model underestimated removal. Sand-bound phages dissociated slowly into the liquid phase, with a detachment constant kdet of 2.6 × 10⁻⁵. This low kdet suggests that river bank filtration plants should be intermittently operated when viral overload is suspected, e.g. during flooding events or at high water-marks in rivers, in order for viruses to become soil-associated during the periods of standstill. Resuming filtration will allow only a very slow virus release from the soil.