The sequential probability ratio test (SPRT), developed by Abraham Wald, is one statistical model available for making mastery decisions during computer-based criterion referenced tests. The predictive validity of the SPRT was empirically investigated with two different and relatively large item pools with heterogeneous item parameters. Graduate students in a course on computer assisted instruction were administered tests on the Dimension Authoring Language and the COM Test, which measured knowledge of how computers functionally work. It was contended that, if the SPRT were used conservatively, it would remain robust as a decision model. Overall agreement coefficients ranged from .84 to .98, depending on the method of determining mastery status on the total test. The Neyman-Pearson classical approach to hypothesis testing was also included. The expected agreement was .95. An average of about 20 test items were required to reach SPRT mastery decisions, reducing testing time by 75 to 80 percent. (GDC)