Fixed and dynamic assemblies of non-spherical particles participate in many heat and mass transfer industrial processes. However, most convective heat transfer coefficients are adapted from correlations available for spherical particles by using the concept of equal volume spheres or in some cases the sphericity, which is a measure of deviation from a spherical shape. In this study, Particle-Resolved Simulations (PRS) are performed to investigate the heat transfer in flow through a fixed random assembly of spherical and ellipsoidal particles of aspect ratio 2.5:1. The assembly of particles is simulated for solid fractions between 0.1 to 0.35 using 191 to 669 particles, respectively, at low to moderate Reynolds numbers (10 <= Re <= 200). The random particle assembly is generated using a physics simulation engine SDK-PhysX. The particle-fluid boundaries are resolved using the Immersed Boundary Method (IBM) with a constant heat flux boundary condition. Simulation results in the ellipsoidal assembly show that the local velocity and temperature field are significantly affected by the particle orientation. Mean Nusselt numbers for the ellipsoidal assembly are consistently larger than spherical particles when Reynolds number is more than 10, and the difference increases as the Reynolds number increases. A Nusselt number correlation is proposed based on the simulation data. The proposed correlation is valid in the range 10 <= Re <= 50 for solid fraction of 0 < phi <= 0.35, and 50 < Re <= 200 for solid fraction of 0.1 <= phi <= 0.35. (C) 2017 Elsevier Ltd. All rights reserved.