The note is concerned with the statistical method applied to numismatic studies with particular reference to a previously published study of the trace-element composition of Augustan coins (G. F. Carter, 1978). It is shown that the grouping procedure proposed by Carter is likely to produce too many significant groups. The compositional differences between predefined groups based on issue and year of manufacture are also considered. Multivariate techniques are used to investigate these differences, and the advantages of this approach in comparison to the inspection of univariate plots are pointed out. -- AATA