These experiments conceptualized stereotype threat as a source of measurement bias in testing, allowing use of mathematical modeling to separate latent from measured variables. To allow this modeling in Experiment 1, White and minority Dutch high school students completed a multi-faceted test that was described as either pertaining to their interests and abilities (control), or as an "intelligence test" (stereotype threat for minority students). In addition, students in the stereotype threat condition completed a form regarding their ethnicity before taking the exam, whereas students in the control condition did so after taking the exam. Modeling of the data showed that stereotype threat produced measurement bias in the minority group, and this bias arose in performance on the most difficult of the subtests. Experiment 2 involved a reanalysis of data from O'Brien and Crandall (2003), illustrating that stereotype threat for men and women in math could be successfully modeled. Experiment 3 involved a multi-faceted math test completed by male and female Dutch undergraduates with no mention of sex differences (control), the negation of sex differences (low stereotype threat), or the endorsement of sex differences in math (stereotype threat for women). Modeling identified that stereotype threat affected math test performance, and the strongest stereotype threat effects again emerged on the most difficult subtest. These studies provide a tool that may be useful for examining stereotype threat in both laboratory and real-world environments in which traditional statistical analyses might be problematic or even impossible to conduct.