The purpose of the study is to investigate the predictive validity of criterion- and norm-referenced grades and the Swedish Scholastic Aptitude Test (SweSAT) and, in particular, possible differences in the prediction of achievement in higher education across academic programs. The analyses were based on credit points obtained by 164,106 Swedish students during the years 1993 to 2001. Two-level modeling with randomly varying slopes with academic program as cluster variable was used. The results provide means and variances of the slopes across the different programs. Variability in the slopes because of program subject area was also investigated. The results indicate that the validity of grades, irrespective of grading system, is stronger in comparison with SweSAT scores. The results also indicate considerable differences in predictive power across programs for the SweSAT, whereas there are much smaller differences for norm-referenced grades and relatively modest differences for criterion-referenced grades. The impact of program subject area on the variability of prediction was substantial for SweSAT scores. (Contains 5 tables.)