How to determine, is a proposed model transformation correct, or not? In general, the answer may depend on the model semantics. Of course, a model transformation is “correct”, if we can extend it to a “correct” instance data transformation. Where should model semantics be defined? Assume, model syntax and semantics are defined in the same meta-model. Then, how to separate syntax from semantics? The answer could be the definition of model schemas proposed in the paper.