The basis, strengths and limitations of Reynold-averaged Navier-Stokes (RANS) and large eddy simulation (LES) turbulence models for computational fluid dynamic (CFD) simulations of industrial gas-fired combustion systems were assessed. Measurement and simulation results from the literature were compared. Applications included a laboratory methane-air diffusion flame, gas flow in a chemical process tube, heat transfer in a 150 kWth gas-fired furnace with a swirl burner, 250 kWth, 1 MWth and 3.6 MWth solid fuel combustion furnaces with swirl burners, and three multi-point ground flares. Results suggested that the roughly order-of-magnitude more computationally expensive LES models were best used to evaluate situations where ignition, flame stability and transient operations were important, such as with flare combustion. The benefits of LES relative to RANS for full-scale industrial furnaces operating in a controlled environment still have not been fully demonstrated. Amongst the RANS model variants, the k – model was favored for use in large, geometrically complex furnaces due to its computational efficiency and ability to represent turbulence over the wide range of velocity scales and flow conditions. For additional computational cost, the realizable k – model provides improved predictions for cases where jet impingement, separating flows, swirling flows, secondary flows and round jet spread are dominant. Due to the complex physics of turbulent reacting flows and complex geometries in industrial furnaces, a specific set of CFD submodels best used to model these applications has not been standardized.