It is demonstrated that the index of dissimilarity ( D), the most commonly used measure of residential segregation, is generally sensitive to the effects of population composition. D is independent of proportion black in an extreme case considered by Duncan and Duncan, but is a logarithmic function of proportion black in two other extreme cases. In the general case, D is in most instances sensitive to compositional changes. It is suggested that the nature of residential succession, in particular the relative stability of the white and black population distributions, influences changes in D. The trend in segregation in Cleveland, Ohio, between 1910 and 1970 is examined in the light of these findings. The properties of a range of other segregation indices are reexamined, and compared with those of D.