The homogenization theory of mammoth silos is investigated in order to establish the homogenization efficiency of mammoth silos. These silos, where homogenization is achieved by intersecting multiple layers during reclaiming of the silo, e.g., by inclining the screw conveyor, can be used for large-scale homogenization of both cohesive and free-flowing materials and are therefore an alternative for blending piles. The presented homogenization model and the calculation of the homogenization efficiency in mammoth silos depend on two variables: the volume distribution and the input properties of the bulk material to be homogenized. The silo geometry and the chosen stacking and reclaiming method determine the first variable. The second variable, time series representing the input properties of a material flow, depends on the material to be homogenized. This paper focuses on modeling input properties and shows that higher order ARMA(p,q) models are required for describing these input properties, instead of the frequently assumed AR(l) models in literature. It does not concentrate on the comparison of predicted and simulated output variances. Conducted simulations of the homogenization efficiency with both ARMA and AR(1) models are found to be very encouraging because the standard deviation of the output properties is reduced on average by a factor 5, i.e., the standard deviation of the output properties is reduced to 20% of the standard deviation of the input properties. (C) 2003 Elsevier Science BY. All rights reserved.