In Model-Driven Engineering bidirectional model transformations emerged as an important ingredient to cope with scenarios such as change propagation, synchronization and to keep consistent system views whenever changes occurring on some view have to be propagated over the others. However, bidirectional mappings open a number of intricate issues that have been only partially solved by research. This master thesis identifies a set of features characterizing bidirectional transformations and validates them against two existing approaches. In particular, a benchmark based on the UML2RDBMS transformation and consisting of two different configurations is implemented by means of two different approaches, such as Triple Graph Grammars and the Janus Transformation Language, for understanding bidirectional transformations with respect to the elicited features.