Particle resolved direct numerical simulation (PR-DNS) has emerged as a promising method to improve gas-particle heat transfer closure models. To date, this method has been applied in random and regular particle assemblies at comparably high void fractions. This paper presents a new methodology for deriving heat transfer Correlations from PR-DNS of very dense particle packings relthiant for packed bed applications. First particle packings were generated using the discrete element method (DEM). After geometric modifications in regions of close particle-particle proximity, a fine mesh with low cell skewness was created for PR-DNS. Grid independence and the effect of the geometry modification were thoroughly investigated. It was also established that steady state simulations are accurate for PR-DNS in this case. Simulations carried out in different assemblies of 100 particles showed significant variation of local transfer rates, implying that it is important to specify a confidence interval when reporting correlations derived from PR-DNS. A newly developed Nusselt number correlation predicts values in the lower range of predictions froth literature correlations. This implies that the use of the currently available correlations may over-predict heat transfer in densely packed beds.