Based on concentration values for ten chemical components of nineteen fragments of Romanglasses found in Norway, three different multivariate classification procedures have been tested. One of the methods is a reiterative point cluster searching system for multidimensional observation spaces, the second method is a conventional eigenvector analysis, and the third method is a disjoint principal component learning machine and classification system. All three methods indicate that there are two main classes of objects in the collection, and the principal component analysis demonstrates that within one of the classes, one sample may be regarded as an outlier -- not detected by the other methods -- possibly representing a third class.