The basic difficulty in the interpretation of eigenfunctions is that our concept of'structure' is weak and that our normative expectations of structure are not clear. The interpretations in such analyses might best look for a low connectivity common structure, the direction of the sign to distinguish nodes, and the appearance of linear sub-systems. The alternative interpretations of Gould's method is merely a matter of personal preference. I AM grateful for Dr Hay's interest in the methods discussed in my paper in Transactions (Tinkler, 1972). Leaving aside the fact that we are not dealing with factor analysis but principal components analysis, I only offered the alternative method using the Ca matrix as one possibility, and one designed to produce a more conventional type of principal components analysis. I did suggest the additional evaluation of both, and possibly other, methods, although Hay's comments appear to be based on the Ugandan examples. I would like to discuss some of the technical issues that he raises, since I believe that they are largely irrelevant and in part inconsistent. Two interpretations of two methods may be involved in so far as the analysis is concerned, but it may well be that we are both mistaken in so far as the analysis of structure is concerned. I will try to answer his points in the order in which they arise. References to Figures will be to those in my original paper. We seem to be agreed that the difficulties of interpreting present/ absent and absent/present in contingency table coefficients is such that we shall ignore them. We are left with two alternatives: to deal with present/present and absent/absent frequencies in a single coefficient, assigning each an equal weight, or to consider simply one of the pairings. In the latter case, we preferentially consider common presences; common absences would yield a similar but inverse picture. In the former case, I can see no reason to favour (conceptually or operationally via a weighting in the coefficient) common presences over common absences. Both it seems to me carry an identical type of structural information. Hay's point that different patterns in the contingency tables yield identical coefficients is in accord with this notion. It mirrors the fact in conventional correlation that two different pairs of variables can yield the same correlation coefficient, where it is not regarded as a disadvantage. I quite agree that the first eigenvector reflects a low level of connectivity component. I had supposed that my discussion of the inverse relationship of high loads to the degree of the modes implied this. This is precisely what one would expect. The relationship is more or less inverted in the direct analysis. Turning to the second eigenvector, I do not see how, if the first component reflects a low level of connectivity amongst the nodes-many common absences-this same view can be extended to this or other eigenvectors. It was my understanding that orthogonalization procedures applied to data matrices were designed to eliminate this very characteristic. I do not see how Hay substantiates his view, a purely negative one, that all the Masindi and the Kampala sub-systems have in common is their lack of contact with Fort Portal and Mbarara. While this is