The results of extensive predictive studies using a mathematical model of pulverised coal combustion are presented. Predictions for two swirl burner types are compared with data collected in a large scale combustor expressly for validation purposes. One of the burner types is reasonably well simulated, while the performance of the other is less well reproduced. A wide ranging study to explore the sensitivities of the predictions to variations of the model parameters was undertaken in an attempt to elucidate the sources of the discrepancies. The difficulty of this detective work was amplified due to the non-existence of reliable data for the velocities of the two phases in a dense pulverised coal flow field. The deficiency in the prediction of the aerodynamic field of the ＇difficult＇ burner is believed to be the principal cause of the prediction problems. This results from the shortcomings of the turbulence model employed for the calculation of strongly swirling two-phase combusting flows.