The conventional method of characterizing the chemical composition of fly ashes is by bulk X-ray fluorescence (XRF). Various indices of reactivity are then calculated from this set of elemental data, for example, the sum of aluminum oxide + silicon dioxide + iron(III) oxide for ASTM C618. The drawback of this approach is that it does not provide an estimate of the uncertainties in the index. For example, the C618 index of a subbituminous coal fly ash determined by bulk XRF is 74.5 %. This is greater than the boundary of 70 % between Class C and Class F ash, and hence it would be assigned to Class F. The mean value calculated from the surface-weighted average of 10,000 glass particles analyzed for chemical composition by automated scanning electron microscopy (ASEM) is 57.8 % ± 18.3 %, which is toward the lower limit of 50 % for the Class C region. Moreover, the frequency distribution is very non-Gaussian and skewed toward the lower values so that the median value is 53 %, which is nearly sub-Class C. This large variance makes any classification system using an index based on bulk composition meaningless. To reduce this variance, the total population of glass particle compositions was divided into a set of clusters, each with a mean chemical composition and a specified band of uncertainty. These clusters typically were observed in glass particle data and presumably were related to the chemical compositions of clay minerals in the coal that was the source of the fly ash. A classification system then could be developed as a set of standard glass compositions with known reactivities. Future research for the development of a standard classification system includes the number of standard glasses, their reactivities, and specifications for the ASEM procedure.