The finite-volume based commercial CFD-code Fluent was used to simulate the reacting flow in a heavy fuel oil fired laboratory furnace. Both the standard k - epsilon turbulence model and the Reynolds stress model (RSM) were tested. The combustion model was based on the conserved scalar (mixture fraction) and prescribed probability density function approach. The heavy fuel oil droplet trajectories were predicted by solving the momentum equations for the droplets using the Lagrangian treatment. The soot distribution in the furnace was calculated by solving a transport equation for the soot mass fraction. Simple expressions for the soot formation and oxidation rates were employed. The radiation heat transfer equation was solved using the finite volume method. The formation of thermal NO from molecular nitrogen was modeled according to the extended Zeldovich mechanism. Fuel-based NO was modeled assuming that all the nitrogen in the fuel is released as hydrogen cyanide (HCN), which then further reacts forming nitric oxide NO or molecular nitrogen N-2, depending on the local combustion conditions. The formation of prompt NO was also included in the calculations. The CFD-code was validated against experimental data for a combustor fired by an industry-type swirl burner for which the initial conditions of the spray have been characterized. It was found that the standard k - epsilon model does not satisfactorily predict the highly swirling flow field in the furnace. The RSM was able to improve the prediction of the flow field. The predicted gas species concentrations were found to be in a reasonable agreement with the measurements, except near the burner and in the vicinity of the furnace axis where discrepancies were found. (C) 2004 Elsevier Ltd. All rights reserved.