Language-centric methodologies, triggered by the success of Domain Specific Languages, rely on precise specifications of modeling languages. While the definition of the abstract syntax is standardized by the 4-layer metamodel architecture of the OMG, most language specifications are held informally for the description of the semantics and the (graphical) concrete syntax. This paper is tackling the problem of specifying the concrete syntax of a language in a formal and non-ambiguous way. We propose to define the concrete syntax by an extension of the already existing metamodel of the abstract syntax, which describes the concepts of the language, with a second layer describing the graphical representation of concepts by visual elements. In addition, an intermediate layer defines how elements of both layers are related to each other. Unlike similar approaches that became the basis of some CASE tools, the intermediate layer is not a pure mapping from abstract to concrete syntax but connects both layers in a flexible, declarative way. We illustrate our approach with a simplified form of statecharts.