The thermal cook–off response of energetic materials (ignition resulting from direct, bulk thermal heating) is important from a safety point of view, but also challenges our understanding of these materials. Explosives are not designed to be cooked off, and, especially in the case of slow cook–off, by the time the material ignites it is substantially different, both chemically and physically, from its original state. In attempting to model such a process numerically, it has generally been assumed that combustion proceeds, from an ignition point, in a more or less planar manner, as has been observed many times in pristine energetic materials at room temperature. To investigate directly the spread of reaction following cook–off in one such energetic material (PBX 9501), small discs of PBX 9501 were heated with a confining glass or sapphire window through which the early stages of the combustion process could be observed directly by high–speed photography. The resulting combustion was found to vary with temperature of ignition, but in all cases was quite different to the laminar burn model. The results of these tests are presented, together with some possible explanations of the behaviour and discussion of the implications to modelling this response.