The determination of local motor vehicle surface temperatures in pre-prototype development has for many years been a critical problem. It is often difficult using numerical models to understand whether limitations lie in the modelling assumptions, or knowledge of the boundary conditions existing in vehicle test conditions. In this paper, we use numerical and experimental techniques to investigate possible accuracy obtainable with RANS-based CFD modeling techniques for local temperature prediction using a well-controlled thermal experiment representative of an underbody situation, operated over several air speeds, exhaust temperatures, and heat shield properties. We will examine the physical phenomena which lie behind observed temperature results using results from tests, RANS and DES simulation, and understand the current best level of correlation obtainable by RANS modeling.