In this dissertation, I explore the systematic failure of the current state of the art statistical techniques to detect gender salary inequity in a special case to propose a more appropriate quantitative method for analyzing gender salary discrimination. This research contributes in three key areas for the development of the quantitative analysis of salary inequity detection. I uncovered salary inequities within gender groups that can mask the salary discrimination between these groups. I then proposed the Two-stage Classification Regression as an appropriate novel statistical method. Finally, the additional propositions made can enhance future salary inequity research.Regardless of the outcome of any gender salary inequity study, we can often find a subgroup of females that is discriminated against when compared to the rest of females. Likewise, a subgroup of males may also be victim of salary inequity when compared to other males. In this context, the first main discovery is that the existence of salary inequities within gender groups can prevent regular statistical techniques from detecting salary inequity between males and females. Detecting this form of salary inequity will increase the sensitivity of the statistical test and hedge its potentially higher risk to the institution.Facing such a statistical problem, the second main contribution was devising a novel statistical approach that can not only succeed where other techniques systematically fail, but also provide a new framework for a more informative statistical analysis. In addition, a more comprehensive definition of salary inequity that goes beyond the simple measure of gender salary gap was derived.The third significant contribution is a set of propositions aiming at framing the agenda for future research on salary inequity studies. A statistical test was proposed to determine when the outcomes of these the linear regression and reverse regression techniques can be expected to be the same. Also, the probability model which is not estimable, but the most robust model was shown to be equivalent to the logistic regression model which is easily estimable, but somewhat difficult to interpret. The goal is to create theoretical supports for better statistical and econometric analyses.