In the development of polymer nanocomposites (PNCs), computational prediction of material properties is of great advantages in achieving rational and cost-effective design. This paper introduces a new multiscale hierarchical approach specifically tailored for PNCs property prediction. Due to its distinct features of generality and comprehensiveness, the introduced methodology outperforms most existing computational methods in this area. By this methodology, a wide range of PNCs containing any-shaped nanoparticles dispersed in a thermoplastic or thermoset polymer matrix can be thoroughly investigated for multiple property prediction. Furthermore, a reliable prediction is promoted by a systematic consideration of the intrinsic multiscale hierarchical structure of PNCs. Most importantly, an effective integration of different methods and theories at individual scales offers unique opportunities for gaining in-depth scientific understandings on the molecular structural origin of macroscopic property enhancement.