The prediction of academic success of undergraduate students using an item response theory partial credit model adjusted grade point average (IRTGPA) is presented and its results are compared to those from an unweighted grade point average (GPA). The sample included 2,444 freshmen admitted in the fall of 1987 to 681 courses at a large midwestern state university. For each student the individual courses, course grades, terms of enrollment, American college testing scores, ethnicity, and gender information were obtained. Analysis focused on three domains: social sciences, liberal arts, and natural and physical sciences. The calculation of an IRTGPA for freshmen and cumulative analyses enhanced the proportion of variance accounted for in the regression models for students in general, men and women as separate groups, and African American and White students as separate groups for the three academic domains. The differences found for students in general, men and women, and ethnic groups indicates that separation of courses into domain-specific groups and the calculation of an IRTGPA enhances prediction of academic success for underrepresented groups in particular. (Contains six tables and nine references.) (SLD)