It has been known since the early sixties that nickel sulfide inclusions cause spontaneous fracture of toughened (thermally tempered) glass, but despite the considerable amount of work done on this problem in the last four decades, failures still occur in the field with regularity. In this study we have classified (by viewing through a 60x optical microscope) inclusions into two groups, which are "classic" and "atypical" nickel sulfides. The "classics" look like the nickel sulfide inclusions found at the initiation-of-fracture of windows that have broken spontaneously. We have compared the structure and composition of the "atypical" inclusions with the structure and composition of the "classics". All of the "classic" and "atypical" nickel sulfide inclusions studied in this work were found to have a composition in the range of Ni52S48 to Ni48S52. Inclusions on the nickel rich side of stoichiometric NiS were found to be two-phase assemblies, and inclusions on the sulphur rich side of NiS were single phase. It had been proposed that the "atypicals" were passive, and of a different composition to the "classics". However, we found that the difference between passive and dangerous nickel sulfide inclusions was not a difference in composition but rather a difference in the type of material in the internal pore space. The passive's had carbon char in their internal pore space, whereas the pore space of dangerous inclusions contained Na2O. The presence of Na2O and carbon char with the inclusions indicates that the formation of the inclusions results from a reaction of a nickel-rich phase with sodium sulphate and carbon.