Employment discrimination is a form of illegal discrimination in the workplace based on legally protected characteristics. In the U.S., federal anti-discrimination law prohibits discrimination by employers against employees based on age, race, gender, sex (including pregnancy, sexual orientation, and gender identity), religion, national origin, and physical or mental disability. State and local laws often protect additional characteristics such as marital status, veteran status and caregiver/familial status. Earnings differentials or occupational differentiation—where differences in pay come from differences in qualifications or responsibilities—should not be confused with employment discrimination. Discrimination can be intended and involve disparate treatment of a group or be unintended, yet create disparate impact for a group.

In neoclassical economics theory, labor market discrimination is defined as the different treatment of two equally qualified individuals on account of their gender, race, disability, religion, etc. Discrimination is harmful since it affects the economic outcomes of equally productive workers directly and indirectly through feedback effects. Darity and Mason [1998] summarise that the standard approach used in identifying employment discrimination is to isolate group productivity differences (education, work experience). Differences in outcomes (such as earnings, job placement) that cannot be attributed to worker qualifications are attributed to discriminatory treatment.

In the non-neoclassical view, discrimination is the main source of inequality in the labor market and is seen in the persistent gender and racial earnings disparity in the U.S. Non-neoclassical economists define discrimination more broadly than neoclassical economists. For example, the feminist economist Deborah Figart (1997) defines labor market discrimination as "a multi-dimensional interaction of economic, social, political, and cultural forces in both the workplace and the family, resulting in different outcomes involving pay, employment, and status." That is, discrimination is not only about measurable outcomes but also about unquantifiable consequences. It is important to note that the process is as important as the outcomes. Furthermore, gender norms are embedded in labor markets and shape employer preferences as well worker preferences; therefore, it is not easy to separate discrimination from productivity-related inequality.

Although labor market inequalities have declined after the U.S. Civil Rights Act of 1964, the movement towards equality has slowed down after the mid-1970s, especially more in gender terms than racial terms. The key issue in the debate on employment discrimination is the persistence of discrimination, namely, why discrimination persists in a capitalist economy.

Gender earnings gap or the concentration of men and women workers in different occupations or industries in and of itself is not evidence of discrimination. Therefore, empirical studies seek to identify the extent to which earnings differentials are due to worker qualification differences. Many studies find that qualification differences do not explain more than a portion of the earnings differences. The portion of the earnings gap that cannot be explained by qualifications is then attributed by some^[who?] to discrimination. One prominent formal procedure for identifying the explained and unexplained portions of the gender wage differentials or wage gap is the Oaxaca–Blinder decomposition procedure.

Another type of statistical evidence of discrimination is gathered by focusing on homogeneous groups. This approach has the advantage of studying economic outcomes of groups with very similar qualifications.

In a well-known longitudinal study, the University of Michigan Law School (U.S.A.) graduates were surveyed between 1987 and 1993, and later between 1994 and 2000 to measure the changes in the wage gap. The group was intentionally chosen to have very similar characteristics. Although the gap in earnings between men and women was very small immediately after graduation, it widened in 15 years to the point that women earned 60 percent of what men earned. From the abstract: Sex differences in hours worked have increased over time and explain more of the sex-based earnings gap, while sex differences in job settings and years spent in private practice have declined and explain less of the gap.

Other studies on relatively homogeneous group of college graduates produced a similar unexplained gap, even for the highly educated women, such as Harvard MBAs in the United States. One such study focused on gender wage differences in 1985 between the college graduates. The graduates were chosen from the ones who earned their degree one or two years earlier. The researchers took college major, GPA (grade point average) and the educational institution the graduates attended into consideration. Yet, even after these factors were accounted for, there remained a 10-15 percent pay gap based on gender. Another study based on a 1993 survey of all college graduates had similar results for black and white women regarding gender differences in earnings. Both black women and white women made less money compared to white, non-Hispanic men. However, the results of earnings were mixed for Hispanic and Asian women when their earnings were compared to white, non-Hispanic men. A 2006 study looked at Harvard graduates. The researchers also controlled for educational performance such as GPA, SAT scores and college major, as well as time out of work and current occupation. The results showed 30 percent of the wage gap was unexplained. Therefore, although not all of the unexplained gaps attribute to discrimination, the results of the studies signal gender discrimination, even if these women are highly educated. Human capitalists argue that measurement and data problems contribute to this unexplained gap.