What do metamodels really look like?

Model-Driven Engineering promotes the use of tailor-made modelling languages for software and systems engineering problems, with metamodels that encapsulate domain knowledge. Despite the importance of metamodelling in MDE, there is little empirical analysis of metamod- els. What are the common characteristics of metamodels? Do modellers follow best practices? How do metamodels evolve over time? How does the size and structure of a metamodel affect the models that conform to it? This paper takes a first step towards answering these questions by automatically analysing the structural characteristics of a corpus of more than 500 publicly available Ecore metamodels.