Model Driven Development may attain substantial productivity gains by exploiting a high level of reuse, across the projects of a same organization or public model repositories. For reuse to take place, developers must be able to perform effective searches across vast collections of models, locate model fragments of potential interest, evaluate the usefulness of the retrieved artifacts and eventually incorporate them in their projects. Given the variety of Web modeling languages, from general purpose to domain specific, from computation independent to platform independent, it is important to implement a search framework capable of harnessing the power of models and of flexibly adapting to the syntax and semantics of the modeling language. In this paper, we explore the use of graph-based similarity search as a tool for expressing queries over model repositories, uniformly represented as collections of labeled graphs. We discuss how the search approach can be parametrized and the impact of the parameters on the perceived quality of the search results.