One of the key challenges in the development of model transformations is the resolution of recurring semantic and syntactic heterogeneities. Thus, we provide a systematic classification of heterogeneities building upon a feature model that makes the interconnections between them explicit. On the basis of this classification, a set of benchmark examples was derived and used to evaluate current approaches to the specification of model transformations. We found, that approaches on the conceptual level lack expressivity whereas execution level approaches lack support for reuse. Moreover, only few of the approaches evaluated provide key features such as an automatic trace model or the ability to reuse specifications by inheritance.