The one-step preparation of sustained release matrix pellets, using a melting procedure in a fluidized bed apparatus, was tested in a 2 3 full factorial design of experiments, using microcrystalline wax as lipophilic binder, theophylline as model drug and talc as additional matrix forming agent. The three influence parameters were (A) size of binder particles, (B) fraction of theophylline in solid particles and (C) fraction of microcrystalline wax in formulation. The response variables were agglomerate size and size distribution, dissolution time, agglomerate crush resistance, sphericity, yield and porosity. Nearly spherical pellets comprising a smooth, closed surface could be obtained with the used method, exhibiting the hollow core typical for the immersion and layering mechanism. The reproducibility was very good concerning all responses. The size of agglomerates is proportional to the size of the binder particles, which serve as cores for pellet formation in the molten state in the fluidized bed. Additionally, the agglomerate size is influenced by the volume of the solid particles in relation to the binder particles, with more solid particles leading to larger agglomerates and vice versa. Dissolution times vary in a very wide range, resulting from the interplay between amount of drug in relation to the meltable matrix substance microcrystalline wax and the non-meltable matrix substance talc. The change of binder particle size does not lead to a structural change of the matrix; both dissolution times and porosity are not significantly altered. Agglomerate crush resistance is low due to the hollow core of the pellets. However, it is significantly increased if the volume fraction of microcrystalline wax in the matrix is high, which means that the matrix is mechanically better stabilized. A theoretical model has been established to quantitatively explain agglomerate growth and very good accordance of the full particle size distributions between predicted and actual values could be shown. A low volumetric binder to solids ratio is compensated by a more porous layer. On the basis of this model, in-depth understanding on the mechanism and influence of product properties could be gained; and an a priori estimation of particle size distributions for new formulas can be performed, with densities, formula, and binder particle size distribution as input parameters.