In analytical solutions, e.g. for evacuation design, the use of computer programs for simulating the smoke spread is common. In recent years a group of computer codes named CFD (Computational Fluid Dynamics) codes has emerged as an engineering tool for describing smoke spread. The CFD codes need to be compared against experimental data so that they can be fully validated. To investigate how different configurations in a retail premises affect the smoke spread and temperatures during a fire, 11 tests were performed. The tests scenario was built in scale 1:2 and can be described as a large room with small ventilation openings near the floor. The configuration parameters were: different fire sizes, different fire positions and different shelf configurations. Heptane pools were used to represent the fires. Three different fire sizes were used and during the test with the largest fire size, 650 mm × 650 mm, the test conditions became under-ventilated, i.e., there was insufficient oxygen available to allow stoichiometric combustion of all evaporated fuel. For the simulations conducted as part of this work, the focus was on under-ventilated fires. However, the experimental results for all of the tests are presented and discussed. The fire tests were simulated using the CFD code FDS (Fire Dynamics Simulator). To see how well FDS simulates under-ventilated fires, both well ventilated and under-ventilated cases were selected for the validation. Gas temperatures and oxygen concentration for the experiments and the simulations, respectively, are compared. Different types of meshes for the simulations and different ways of modelling the fire were used. The results of the validation show that the combustion model (mixture fraction combustion model) with empirical amendments for when the fire is allowed to burn, is very sensitive to changes in the oxygen level. The comparison between the experimental data and the simulation indicates that FDS easily can underestimate the oxygen level and thereby the heat release rate which, in turn, creates an underestimation of the temperatures. The validation has shown that the simple empirical expression used for when the fire is allowed to burn is very sensitive and if used without proper understanding it may produce large differences between the experiments and the simulations. It is also clear that the temperatures for well ventilated cases may be overestimated and that the use of visibility and toxicity (soot and carbon monoxide yields) are related to uncertainties. It should also be noted that there are cases where the temperature from the simulations and the temperature measurements correspond relatively well with each other and yet other cases when the simulated temperature is higher than the measured temperature. This depends on the simulation case, the position in the set-up and the time period compared.