Sampling of particles smaller than 100 nm is of great concern for assessing the adverse health effects of airborne nanoparticles in both atmosphere and workplaces. Inertial filter (IF) is a new application of air filter, which is recently developed by the present authors to classify nanoparticles by passing aerosol through an air filter at a high filtration velocity. In the present work, we investigated the effects of fiber packing density and particle deposit on classification performance of IF to achieve 50% cut-off diameter of 100 nm for sampling nanoparticles. Unwoven stainless steel (SUS) fiber mat with diameter of 8 mu m is used as a filter medium. It was found that, in inertial regime, the collection efficiency changes markedly with the packing density due to the interactions of neighboring fibers. As a result, by adjusting the filtration velocity to achieve 50% cut-off diameter of 100 nm, IF with various packing densities had almost the same classification performance, and a compressed filter is advantageous because it had a lower pressure drop. The influence of loaded particles on the classification performance was also investigated. It was found that, up to the collected particle mass of 1 mg, the effect of dust load is not significant. Therefore IF with a higher packing density is superior in achieving 50% cut-off diameter of 100 nm from the viewpoints of both initial and dust-loaded classification performance.