The problem of determining the optimal cross-reference structure for a given index and for a given community of users is discussed. A cross-reference structure is represented as a graph in which the nodes are index terms and the links are relations between index terms. In order to clarify the concept of 'level of cross-referencing' the characteristics of cross-reference structures are studied. Some measures of cross-references distributions are suggested as a means of comparing the cross-referencing levels of subject indexes. Types of relations linking the terms of cross-references in existing indexes and thesauri are examined. The implications of the study for the construction and testing of indexes and thesauri are discussed. (Author)