Principal components and cluster analyses are sometimes used to classify glass specimens on the basis of their chemical composition. This paper examines a variety of approaches designed to overcome criticisms of the more usual approaches. Different methods can, potentially, give different results and can be unduly, and predictably, influenced by a subset of oxides in a composition. It is argued that separate analyses of the major and minor oxides, as well as any global analysis, will, in consequence, often be sensible.