The paper is concerned with particle-laden turbulent flows and its efficient and reliable simulation. The continuous phase is calculated using a large-eddy simulation technique along with a finite-volume method for block-structured curvilinear grids. The particulate phase is simulated using a Lagrangian approach where hundred thousands of individual monodisperse particles with varying particle diameters are released and tracked throughout the computational domain. To allow such a large number of particles, a highly efficient tracking algorithm is applied, where particle paths are predicted in an orthogonal computational domain, avoiding a time-consuming search algorithm. Both simulation algorithms, for the continuous and particulate phases, are completely vectorized and parallelized using domain decomposition. Two different test cases are considered. First the particle-laden turbulent flow through a 90° bend with tubular cross-section at ReD = 10,000 is studied. The predicted results of aerosol deposition efficiency, over the entire range of particle diameters considered, show an excellent agreement with experimental measurements. Second the aerosol deposition in an idealized mouth geometry with a relatively small inlet diameter is tackled for a steady inhalation flow rate. Particles of different diameters (2.5, 3.7 and 5.0 μm) are released continuously at the inlet and tracked through the flow field. Two different methodologies are investigated using either a frozen LES flow field or a dynamic tracking of the particles. Regarding the total aerosol deposition, the results of both methods are compared with experimental data from the literature and found to be in good agreement.