The numerical simulation of fires concerns an increasingly wide range of applications, among which archaeology. The Chauvet-Pont d’Arc Cave (France) presents thermal marks on some of its walls. These marks result from prehistorical fires and archaeologists would like to know their characteristics (fire location, mass of fuel, number of fires, etc) to make assumptions about their functions. Being impossible to carry out fires in the cave, combustion simulation is a suitable tool to discuss this question. However, simulating a fire in a confined geometry is still challenging and some inherent difficulties must be overcome. In that respect, we reproduced four experimental fires in a cave-like geometry. Four tepee-like hearths, composed of different initial wood masses (16, 32, 60 and 90 kg), were burnt in an underground quarry. Twenty-seven thermocouples measured the temperature in the cold and hot gas layers, as well as in the ceiling jet area. The measurements in the ceiling jet resulting from the 90 kg fire are successfully compared to the corrected Delichatsios correlations. In addition, the measured mass loss rate is compared to a tailored modeling for tepee hearths. We also propose three strategies to simulate the combustion process of the 90 kg hearth: (i) with a significant refinement in the ceiling jet area (approximately 1,500,000 cells) and the law of the wall, (ii) without refinement but with Newton’s law of cooling and (iii) without refinement but with simplifying assumptions and adjustments. While the first and third approaches provide meaningful results, the second case is found to be inconvenient for compartment fires. Based on experimental data, this study aims to give some keys to perform simulations of compartment fires (in tunnels, buildings, car parks, etc) and numerically approximate the combustion of wood.