Many software engineering processes produce series of related models. They describe the designed system from different viewpoints, each emphasizing a particular concern. Models are produced by individual teams and sometimes third party developers, and sometimes they are synthesized using tools. It is unlikely that they are developed within a single modeling repository or share common repository identifiers. To reason about the models, a correspondence relationship needs to be established between model elements representing the same concepts. Current approaches for inferring correspondence use fixed semantics built into the algorithms and derived from small modeling language subsets. This paper presents a flexible approach for inferring correspondence. The rules are generated based on a Meta-Object Facility compliant metamodel specifying modeling language of the compared model instances. They can be refined with additional domain specific rules. The approach can be used as a basis for comparing, merging and reconciling models.