Filtration is commonly employed in water and wastewater treatment to remove particles and reduce the concentration of microbial pathogens. All physical models of packed-bed filtration are based on a proportional relationship between particle removal per unit depth of bed and the local particle concentration, dC/dz = -C/l, where l is the filtration length scale. Although l is known to vary with time and filter depth for heterogeneous suspensions or "dirty" beds, this paper demonstrates that the filtration rates of even seemingly monodisperse particle suspensions under clean-bed filtration conditions cannot be described with a single filtration length scale. A new model is derived for particle filtration that accounts for heterogeneity at two different spatial scales. Heterogeneity at the scale of the pathogen and/or collector (microscale heterogeneity) leads to a slow power-law decay of contaminant concentration with distance, instead of the fast exponential decay predicted by the standard model. Heterogeneity at the filter scale (macroscale heterogeneity) provides another level of complexity that can be evaluated once microscale heterogeneity effects are characterized. This model for microscale and macroscale heterogeneous particle filtration is verified by filtration experiments on a recombinant analogue of the waterborne pathogen Norwalk virus.