This paper argues that the Oaxaca-Blinder decomposition incompletely decomposes the mean wage difference between two comparative groups into a portion that is explainable by differences in characteristics, and a portion that is attributable to differences in the valuation of these characteristics. As a result, studies on earnings inequalities that rely on the traditional methodology, or its other variants, could incorrectly estimate the explained and unexplained components of the wage gaps. To demonstrate the potential problem, this paper proposes an alternative decomposition technique that is an extension of the traditional method. The technique involves estimating the parameters for an assumed investment function for the unobserved post-schooling (on-the-job) training. To empirically illustrate the concept, the proposed methodology is applied to the study on gender wage inequality using Australian 2001 data from the HILDA survey. The traditional Oaxaca-Blinder decomposition is also carried out on the same data set for comparison. The results do indicate very different estimates obtained using the two techniques.