Segregation between ethnic or sociodemographic groups represents a longstanding key independent and dependent variable for the social sciences. However, researchers have only recently begun to take advantage of inferential rather than descriptive statistical techniques in order to assess various aspects of segregation. Specifically, this paper shows that the multilevel binomial response approach suggested by Leckie et al. (2012) provides a particularly flexible framework for describing and explaining segregation in ways not previously possible. Taking the index of dissimilarity (D) as an example we demonstrate how the multilevel binomial response approach helps to reduce the problem of small unit bias, allows to asses segregation at different scales and enables researchers to better understand the role of individual- and contextual-level explanatory variables in shaping segregation. To this end, we employ three case studies focusing on different manifestations of ethnic and gender segregation using survey data from urban, national and cross-national settings.