This paper aims at opening a discussion on the quality assessment of indirect test strategies in the context of analog and RF integrated circuit testing. Many parameters may influence the prediction efficiency of the indirect test model (choice and number of indirect parameters taken into account, learning algorithm used to build the model...). In order to evaluate the quality of a given model, several metrics can be evaluated, that reflect either the average prediction error, a global reliability or a misclassification rate. But what are the most pertinent metrics to reflect the level of confidence that can be expected from the indirect test to efficiently replace a traditional test based on RF measurements? Which metrics can lead to an informed choice of an indirect test strategy for its stability and predictive power? These considerations are investigated in this paper and illustrated in a practical case study.