The efficiency of sensor-based automated sorting depends on both correct identification and separation of different types of particles. It is known that the distribution of particles fed to the sorter will affect both of these. When different particles are in close proximity, they can be "agglomerated" or seen as a single particle during identification and also have an increased probability of being unintentionally co-ejected. Both factors will have a negative effect on separation efficiency. The aim of this work was to model the air ejection manifold of a sensor-based automated sorter and to investigate the relationship between particle proximity and unintentional co-ejections. The airflow from a single air ejection valve of a sorter was modelled using computational fluid dynamics (CFD) software and calibrated against a Tomra Sorting Solutions optical sorter. It was found that the air ejection manifold could be accurately represented in CFD code. Particles were modelled using the discrete element method (DEM) software and the effect of particle position, relative to an air ejection valve, on accurate ejection was examined using an integrated CFD-DEM model. The results of these models matched reasonably well with physical measurements. The models created can be used as a basis for the prediction of sorter efficiency. (C) 2015 Elsevier Ltd. All rights reserved.