Testing every possible combination of the input parameter values is often impractical, inefficient or too expensive. One common alternative is pairwise testing where every pairwise combination of the parameter values is tested. Although pairwise testing significantly reduces the number of test cases, the challenge remains in analyzing the test outputs to discern the precise characteristics of parameters causing the failures. This thesis proposes a novel approach to output analysis by identifying “bad pairs”: pairs that always result in failed test cases. A framework implementing the proposed approach is presented together with three case studies. Results from the case studies suggest there are positive relationships among the numbers of failed test cases, faults, and independent bad pairs. Also, filtering of test cases seems to have a significant impact on the bad pairs identified. We believe the proposed approach can facilitate the debugging process in software testing.