Model transformation can't be directly tested using program techniques. Those have to be adapted to model characteristics. In this paper we focus on one test technique: mutation analysis. This technique aims to qualify a test data set by analyzing the execution results of intentionally faulty program versions. If the degree of qualification is not satisfactory, the test data set has to be improved. In the context of model, this step is currently relatively fastidious and manually performed. We propose an approach based on traceability mechanisms in order to ease the test model set improvement in the mutation analysis process. We illustrate with a benchmark the quick automatic identification of the input model to change. A new model is then created in order to raise the quality of the test data set.