AbstractIn order to determine the degree of correspondence between sets of multivariate observations based on different kinds of traits, two new methods, derived from fundamentally different notions of “correspondence,” are adopted here and compared. Using networks or trees to represent contemporary relationships, the first method tests the similarity of the cluster or hierarchic structures implicit in two sets of data. The second approach tests the departure from perfect geometric congruence or superimposability. Computer simulation was used to generate the distributions needed for significance tests under the null hypothesis.By the first technique, we find significant correspondence among the cluster structures for geographic, allele frequency, and anthropometric data on 19 Yanomama Indian villages. The results are similar and more precise for a subset consisting of seven villages. Some of these results differ from the conclusions which would be reached with the conventional correlations based upon entries in distance tables.The direct test of congruence, used only for the data on the subset of seven villages, gives results which differ substantially from those based on cluster‐structure. There are, however, similarities between the measure of congruence and the simple correlations based on entries in the distance tables.The significant correspondences observed call for some explanation. Cultural and demographic features determine the particular non‐random allocation of individuals to village fragments when a village splits. These social phenomena are invoked in tentative explanation of the agreement among historical, biological, and geographic relationships of villages.