Hubert and Subkoviak (1979) have recently proposed a confirmatory technique which allows testing whether a given proximity matrix or some geometrical representation of it can be said to possess certain predicted structural properties. Their approach is essentially very simple. They compute the correlation between the set of values q(oi, oj), which are either the proximity values for objects oi and oj or a distance value for the respective point-representations, and the collection of c(oi, oi), the numerical assignments given to these pairs on the basis of some theory. Thus, if an observation ofq(o i , o/.) > q(ok, or) were hypothesized, then, for example, c(oi, o i) = 2 > 1 = c(ok, ol) might be defined. Having set up a s t ructure function c in this way, a (linear or rank-order) correlation between c and q is then an index of how well the actual data or distances correspond (linearly or ordinally) to this particular choice of numerical predictions. That is, the theory used to generate the function c(oi, oj) is given empirical support if the two sets o f elements, c(oi, o i) and q ( o i , Oj), have a similar patterning of high and low entries (Hubert and Subkoviak, 1979, p. 363). In order to be able to evaluate the probabili ty with which such a correlation can be expected to occur for random data, it is possible simply to compute the correlations for all or a reasonably large sample of the permutat ions of the sets of c(oi, o i) o r q(oi, o]) values, respectively. If the observed correlation then exceeds some sufficiently high percentage point of the cumulative distribution of the generated set o f