Abstract.A numerical investigation of a bluff-body stabilised nonpremixed flame, and the corresponding nonreacting flow, has been performed with differential Reynolds-stress models (DRSMs). The equilibrium chemistry model is employed and an assumed-shape beta function PDF approach is used to represent the interaction between turbulence and chemistry. The Reynolds flux of the mixture fraction is obtained from a transport equation, hence a full second moment closure is used. To clarify the applicability of the existing DRSMs in this complex flame, several models, including LRR-IP model, JM model, SSG model as well as a modified LRR-IP model, have been applied and evaluated. The existing models, with default values of the coefficients, cannot provide overall satisfactory predictions for this challenging test case. The standard LRR-IP model over predicts the centreline velocity decay rate, and therefore does not perform satisfactory. The modified LRR-IP model, with model constant C 1 = 1.6 instead of the standard value 1.44 (here named BM-M1), gives better results for the mean velocity. However in the nonreacting case this does not lead to improvement in predicting rms fluctuating velocities especially downstream of the recirculation zone. Motivated by the need to improve the prediction, a new modification of the LRR-IP model is proposed (BM-M2), with model constant C 2 = 0.7 in the pressure strain correlation rather than the standard value 0.6. With the new modified model, a very significant improvement of the prediction of flow field is obtained in the nonreacting case, whereas in the reacting case the prediction of the flow field is of the same overall quality as with BM-M1. This shows that some DRSMs have different behaviour in the nonreacting case and the reacting case. In the reacting case also the mean and variance of mixture fraction are considered and it is found that the best results are obtained with the BM-M1 model, with SSG as second best. Combining the results for flow field and mixture fraction field it is concluded that the BM-M1 model is recommended for further studies of this bluff-body stabilised flame. Grid independence of the result is demonstrated.