Purposes of this research were to (1) analyze the results of the grade-point average adjustment using 5 methods namely; Angoffs linear method design IV C-2 (ANGOFF-4C2), IRT grade response model (IRT-GRM), GLM model (GLM-MODEL), one factor congeneric measurement method (CON-CFA) and many-facet Rasch model (RASCH-FACET) (2) compare the predictive validities of the adjusted GPA from the 5 methods when using university GPA in the first and the second year as criteria. The upper secondary school GPA of 5,919 students who graduated in B.E. 2539 were used as predictor. These students of 28 schools were randomly selected, under the jurisdiction of Department of General Education which were grouped into three levels pertaining to their educational quality. Entrance examination score and GPA in the first and the second year of 1,029 of these students who passed the entrance examination for public university in B.E. 2540 were used as criteria. Major findings were as follow: 1. When comparing to adjusted GPA, four types of actual GPA were found: (1) the actual GPA was higher at all ability levels, (2) lower at all ability levels (3) higher at the low ability level and lower at high ability level and (4) higher at the high ability level and lower at the low ability level. 2. When using adjusted university GPA in the first and the second year from regression equation and RASCH-FACET as criteria yielded concurrent results, that is at .05 level of significance adjusted GPA in secondary school from RASCH-FACET, ANGOFF-4C2 and IRT-GRM yielded higher predictive validity than actual GPA, while CON-CFA yielded lower predictive validity than actual GPA. It was found however, that when adjusted university GPA in the first and in the second year from regression equation were used as criteria, GLM-MODEL yielded higher predictive validity significantly at .05 level, than actual secondary school GPA. This difference was not found when adjusted university GPA in the first and the second year from RASCH-FACET were used. Adjusted GPA fromRASCH-FACET yielded the highest predictive validity, followed by ANGOFF-4C2 and IRT-GRM which were equal, and by GLM-MODEL and CON-CFA respectively.