A computational study has been performed to evaluate the predictive capabilities of some existing eddy-viscosity (both linear, LEVM, and non-linear, NLEVM) and Reynolds stress transport turbulence models (RSTM) by reference to a transonic shock-induced separated flow over a 10% axisymmetric bump. The calculations have been carried out during the course of a collaborative research programme including both UK universities and industry. The findings of the project demonstrate that improved results can be obtained for such flows by using more advanced turbulence models. For linear eddy-viscosity models, only the SST approach gave good predictions of shock location, recirculation size and pressure recovery, although this was accompanied by deficiencies in the prediction of post-shock velocity profile shape. Non-linear eddy-viscosity models, particularly at the cubic level, provided a more consistent level of agreement with experiments over the range of shock location, wall pressure and velocity profile parameters. Some improvement was also seen in the prediction of turbulence quantities, although only a move to an RSTM closure model reproduced the measured peak stress levels accurately. It was notable that the use of low-Re variants of the models (instead of wall functions) produced no significant improvement in predictions. There are, however, some shortcomings in all models, particularly in the development of flow after reattachment, which was always predicted to be too slow.