In order to reliably predict the particle number filtration of gasoline particulate filters (GPF) under practical driving conditions, an extension to established filtration models is developed. For the validation of this approach and in order to close a gap of available measurement data at high space velocity in the literature, the particle-size-resolved fresh filtration efficiency of seven different cordierite filters is determined experimentally. Moreover, the experiments on a dynamic engine test bench focus on the impact of the pore-size distribution and the filter wall thickness under steady-state as well as transient, cold-start conditions. In order to model all trends observed, a new correlation for the particle collection due to inertial deposition is proposed and embedded in a heterogeneous multiscale model framework for a GPF. The presented approach can predict all trends observed in the measurements, including a stabilization of the filtration efficiency with increasing space velocities above a certain level. A comparison of several modeling approaches reveals the partly different behaviors at varying space velocities for the here presented model as well as for established filtration models.