Stress autocorrelation functions (SAF) of entangled polymer melts are calculated by coarse-grained molecular dynamics (MD) simulations. We show that time-averaged stresses reduce the strong noise in SAFs while still capturing most relevant relaxations of the chains. Plateau values of the SAF compare well with plateau values predicted from the entanglement length evaluated via primitive path analysis (PPA) and from experiment (where available). Three types of polymer models are studied: a flexible bead spring model, a semiflexible bead spring model, and a coarse grained bisphenol A-polycarbonate (BPAPC) model. This approach provides a straightforward way to analyze rheological properties of polymer melts on the fly during a simulation.