Abstract The dispersion of inorganic particles within polymeric materials is an extensively used method to enhance their mechanical properties. One of the major challenges in the simulation of polymer composites is to model the uneven surface of the fillers which strongly affects the dynamics of the adsorbed polymers and consequently the macroscopic mechanical properties of the final composite. Here we propose a new multiscale approach that, using experimental adsorption data, constructs the filler surface to statistically reproduce the surface defects. We use this approach to analyse the structure and dynamics of highly entangled polyisoprene melt in contact with different realistic carbon black samples. We show that the presence of the heterogeneous surface has a negligible influence on the structure of the polymer chains but a major effect on their dynamics and the surface wettability.