An off-line global three-dimensional tracer model based on analyzed wind fields was augmented to simulate the atmospheric transport of mineral dust. The model describes the evolution of the aerosol size distribution and hence allows to compute aerosol number and mass concentrations. In this study we describe the parameterization of the sedimentation process and include a preliminary source formulation but exclude wet deposition. Validation of the model is done during a 16-day period in June-July 1988 with very scarce precipitation. It is based on a comparison of every model grid box with daily satellite-derived optical thickness observations of Saharan dust plumes over the North Atlantic and the Mediterranean. The model reproduces accurately the daily position of the dust plumes over the ocean, with the exception of Atlantic regions remote from the African coast. By systematic analysis of transport and aerosol components we show that the largest uncertainty in reproducing the position of the dust clouds is the correct localization of the source regions. The model simulation is also very sensitive to the inclusion of convection and to an accurate treatment of the sedimentation process. Only the combination of source activation, rapid transport of dust to higher altitudes by convective updraft and long-range transport allows the simulation of thc dust plumes position. This study shows that a mineral dust transport model is only constrained if both the sourcc strength and the aerosol size distribution are known. The satellite observation of optical thickness over the Mediterranean and assumptions about the size distribution indicate that the dust emission flux was of the order of 17x 106 t for the 16-day period under invcstigation. The simulations suggest that a major aerosol mode initially around 2.5 [tm with a standard deviation of 2.0 plays the dominant role in long-range transport of mineral dust.