Model transformations can automate critical tasks in model-driven development. Thorough validation techniques are required to ensure their correctness. In this lecture we focus on testing model transformations. In particular, we present an approach for systematic selection of input test data. This approach is based on a key characteristic of model transformations: their input domain is formally captured in a metamodel. A major challenge for test generation is that metamodels usually model an infinite set of possible input models for the transformation. We start with a general motivation of the need for specific test selection techniques in the presence of very large and possibly infinite input domains. We also present two existing black-box strategies to systematically select test data: category-partition and combinatorial interaction testing. Then, we detail specific criteria based on metamodel coverage to select data for model transformation testing. We introduce object and model fragments to capture specific structural constraints that should be satisfied by input test data. These fragments are the basis for the definition of coverage criteria and for automatic generation of test data. They also serve to drive the automatic generation of models for testing.