Fires in confined spaces are of major concern in fire safety engineering. Indeed, fires with ensuing fatalities, generally occur in apartment or room fires. The decision and response time of rescue teams mainly results from empirical understanding. However, since compartment fires are multi-physical and multi-scale problems, a clear fundamental approach is needed. One of the main issues concerns the transition from localized to generalized fire. The most important vector leading to generalized fires is smoke. Indeed, smoke temperature is very high and many species i.e. burned and unburned gases are already or might mix. In this study, an experimental facility composed of a maritime container is set-up. It allows enclosure fires up to of power. This power is representative of moderate room fires. The global behavior of the smoke is investigated through the analysis of mean experimental temperature fields, smoke dynamics by large scale PIV as well as numerical simulations. For numerical simulations, the Fire Dynamics Simulator (FDS) software is used. The experimental measurements are used to evaluate the validity of FDS in under-ventilated conditions. It is observed that FDS code is able to reproduce both temperature and velocity data of enclosure fires with accuracy depending on the power to volume ratio. A criterion capable to discriminate the ventilation status of confined fire is observed and discussed.