This thesis describes an experimental investigation of the flame structure close to the extinction and the blow-off events of non-premixed and spray flames stabilized on an axisymmetric bluff body in a confined swirl configuration. The comparison of flames of different canonical types in the same basic aerodynamic field allows insights on the relative blow-off behaviour. The first part of the thesis describes several velocity measurements in non-reacting and reacting flows. The main usefulness of this data is to provide the aerodynamic flow pattern and some discussion on the velocity field and the related recirculation zones. The velocity and turbulence information obtained are particularly useful for providing data, which is crucial for validation of computational models. The second part describes an experimental investigation of non-premixed stable flames very close to the blow-off condition. The measurements included visualisation of the blow-off transient with 5 kHz OH* chemiluminescence, which allowed a quantification of the average duration of the blow-off transient. OH-PLIF images at 5 kHz for flames far from and close to extinction showed that the non-premixed flame intermittently lifts-off the bluff body, with increasing probability as the fuel velocity increases. The flame sheet shows evidence of localised extinctions, which are more pronounced as approaching blow-off. The measurements include blow-off limits and their attempted correlation. It was found that a correlation based on a Damkohler number does a reasonable job at collapsing the dataset. The final part examines the blow-off behaviour of swirling spray flames for two different fuels: n-heptane and n-decane. The measurements include blow-off limits and their att~mpted correlation, visualisation of the blow-off transient with 5 kHz OH* chemiluminescence, and the quantification of the average duration of the blow-off transient. It was found that the average duration of the blow-off event is in order of the tens of ms for both spray flames (10-16 ms). The blow-off event is therefore a relatively slow process for the spray ~ames using n-heptane and decane fuels. This suggests that control measures, such as fast fuel injection, coupled with appropriate detection, such as with chemiluminescence monitoring, may have a reasonable chance of success in keeping the flame alight very close to the blow-off limit. These results, together with those obtained for the non-premixed gaseous case form a wide body of experimental data available for the validation of turbulent flame models. The quantification of some properties during the blow-off transient can assist studies of extinction based on large-eddy simulation that have a promise of capturing combustion transients.