Proposals for multivariate analysis with ordinal variables based on analogies with product-moment formulae are criticized on the grounds that such analogies do not provide a basis for interpreting the resulting statistics in terms of evidence bearing on an hypothesis concerning the presence or absence of some specified relation in the data. Thus, it is concluded that the long-standing problem of multivariate analysis with ordinal data still awaits a satisfactory solution. It is suggested further that the fundamental difficulty obstructing work on the problem is the lack of clear, precise formulation of what constitutes a multivariate ordinal relation.