Building upon the ideas first exposed by Theil and Finizza (1971) and Fuchs (1975), this paper presents an additively decomposable segregation index based on the entropy concept used in information theory. For any pair of classification variables in a given year, the index is decomposed into a between-group and a within-group term. To analyze intertemporal changes in gender segregation for a given partition, the index is decomposed into two terms that capture, respectively, gender composition effects, and changes in the groups' demographic importance. These decompositions are illustrated with Spanish data on occupations and human capital levels for 1977 and 1992. It is found that, in both years, the higher the educational level, the smaller is gender segregation for most age groups. Moreover, gender segregation decreases with age in all educational categories. However, most gender segregation takes place within, rather than between, age/education categories. Lastly, changes in gender composition across occupations, nearly offset by occupational mix changes, account for a decline of 2% in total gender segregation over this period.