The concept of heterogeneity is very important in XML data management, since many common applications must deal with large and complex collections which do not conform to a schema. Heterogeneity in XML collections can be present at many different levels (textual and structural) and needs to be addressed from several perspectives. This paper contributes a formal characterization of heterogeneity in XML collections based on information-theoretic considerations. We show how it can be applied in some important use cases, and we demonstrate its effectiveness by using it to analyze a number of relevant XML collections and retrieval approaches found in the literature. We show that a large space of highly heterogeneous collections has not??been adequately addressed by these approaches.