Belt finishing is a superfinishing process that enables to reach fine surface roughness. This paper proposes a numerical model, predicting belt finishing roughness profiles of a 27MnCr5 carburized steel. It utilizes belt topographies that are then numerically deformed thanks to a specific morphological filter. A method for determining the morphological filter parameters is proposed and depends on the size of the grains. Then, the deformed topography of the belt virtually scratches the workpiece at a given depth of indentation (computed by a rough contact model). The model predicts surface roughness profiles in the steady state regime. The paper shows a reasonable agreement between experimental and numerical results. The model captures the main trends: the higher the grain size the higher the roughness of the workpiece. Computational time needed for this model is a few minutes, making it a useable solution for the industry.