Although fault injection is a powerful technique to exploit implementation weaknesses, this is not without limitations. An important preliminary step, based on rigorous calibration of the fault injection equipment, greatly affects the accuracy and repeatability of the results. Even if equipment calibration can be performed directly with the target application, it can be more relevant to use fault characterization tests instead, which are small programs designed to propagate fault effects, in order to identify more easily faulty behaviors. However, designing tests that propagate most fault effects is not an easy task. In this study, we propose original metrics to evaluate fault characterization tests based on fault propagation and fault discrimination. These metrics are used to compare the performance of different fault characterization tests from the literature, so as to identify the most important design choices according to a wide range of fault models. From these observations, we detail a general approach to design more efficient tests. Finally, we propose a minimal test set based on these guidelines, to quickly identify settings with high fault probability on a Cortex-M4 32-bits microcontroller, using voltage fault injection.