The particle capture process of fibrous filter depends on not only filtration velocity, packing density of fibers, particle and fiber diameters, but also fiber arrangement (e.g. parallel model, staggered model, and separation ratio). In this work, the lattice Boltzmann-Cellular Automata (LB-CA) probabilistic model for particle-laden flows was introduced to simulate the filtration process of multi-layer fiber filters with different fiber arrangement, and then the effects of the fiber arrangement on the capture efficiency, pressure drop, quality factor, and capture contribution were investigated. It was found that the staggered model performs better than the parallel model and its capture efficiency is higher than that of the parallel model. In addition, the capture efficiency and pressure drop both increase with the fiber separation ratio, while the quality factor decreases. With respect to the capture contributions of various rows of fibers in the steady and unsteady filtration processes, the fore fibers (especially the first row in the parallel model and the first two rows in the staggered model) play a more important role than the rear ones, especially when inertial impaction is predominant. Furthermore, the dynamic evolution of particle deposition patterns, pressure drop, capture efficiency, quality factor, and capture contribution due to different capture mechanisms in the particle loading of multi-fiber filters were investigated. The key factors to determine the dendritic structures are the shielding effect in the parallel model and the "river diversion" effect in the staggered model. Both the capture efficiency and pressure drop increases with the particle loading, however the pressure drop increases faster, resulting in the decrease of the quality factor. For the loading process of larger particles in the parallel model the pressure drop from the LB-CA model agree well with the predictions of the empirical model, and the interceptive capture efficiency predicted by the empirical model is also in good agreement with the LB-CA results. However, no model considers the fiber arrangement and different capture mechanisms. The LB-CA simulation provides good understanding in the dynamic filtration processes of multi-layer fibrous fibers, which is useful for filter design and arrangement optimization. Crown Copyright (C) 2013 Published by Elsevier Ltd. All rights reserved.