The development of a valid measure of academic performance through the application of existing statistical methodology in a new manner was studied. Focus was on analyzing the predictive validity of traditional pre-admissions measures in determining overall academic performance in college. This new criterion, termed a Universal Scale for Grades, is used to determine the predictive validity of certain pre-admissions measures for a particular cohort of students at one university. Two item response theory (IRT) models--G. N. Masters' (1982) Partial Credit Model, and F. Samejima's (1969) Graded Response Model--were used in this study. As far as known, this study represents the first application of IRT methodology to the problem of equating grades from different courses to a common scale. Data were taken from the undergraduate admissions office and the registrar's office at Stanford University. The sample consisted of all 1,564 students entering Stanford as freshmen in the autumn quarter of 1982. The database consisted of verbal and mathematics scores on the Scholastic Aptitude Test, College Board Achievement Test scores, high school grade point averages, rank in and size of high school class, and admissions staff assigned academic rating and non-academic rating. Only those records in course files that contained a letter grade were used. Principal factor analysis and validation results indicate that use of the IRT models to create more reliable measures of academic performance shows promise. (TJH)