Premixing of fuel oil and air for the purpose of lean-premixed combustion in gas turbine combustors was studied experimentally with laser-induced fluorescence (LIF) and numerically using a modified KIVA-3 code. A full-scale industrial burner, operated at atmospheric pressure, was investigated as well as a test burner of reduced scale, operated at pressures up to 15 bar. Three improved-accuracy models were used in the simulations to replace the corresponding standard KIVA-3 models: (1) a droplet-vaporization model including multiple components and real-gas effects; (2) a droplet-dispersion model, using a linear filter; and (3) a swirl correction to the k-epsilon turbulence model. As upstream boundary condition for the spray simulation, drop-size distributions measured under cold-spray conditions were used with Sauter mean diameter (SMD) correlated according to Lefebvre. Calculations with different spray models were performed to study the effect of multicomponent vaporization, droplet-internal inhomogeneities, property-estimation methods, dispersion modeling, and swirl correction to the k-epsilon model. Considerable differences were found in fuel-vapor distribution when using different models. The LIF measurements and the numerical simulations were compared in terms of the fuel-vapor distribution. The simulation results were found to be very sensitive to the initial SMD, in particular in the presence of air swirl. Inaccuracies in the estimated SMD, therefore, constitute the primary accuracy limitation of the simulation. By adjusting the SMD, in many cases good agreement between measurement and simulation could be obtained. Remaining discrepancies that cannot be attributed to SMD adjustments are supposed to originate either from a turbulence-modeling error and the occurrence of secondary droplet breakup, which was not modeled, or, on the other hand, to the error by assuming a proportionality between LIF intensity and fuel oil concentration. (C) 2001 by The Combustion Institute.