the aim of this study was to identify upland rice genotypes with high stability and adaptability by the gge biplot method based on the predicted genotypic and phenotypic values. of the 20 genotypes evaluated, 14 were lines developed by the cooperative program for rice improvement of minas gerais and six were controls. the gge biplot analysis showed that cultivar brs pepita and mg1097 were closest to the ideal genotype. in the comparison of the fixed with the random models (% g + ge, prediction error sum of squares and correlation), it was observed that the use of phenotypic means in all comparative parameters indicated a lower predictive potential under simulated imbalance than the use of predicted genotypic values. the conclusion was drawn that brs pepita and mg1097 are ideal genotypes for southern minas gerais and that the predictive power of the phenotypic means underlying the study of stability and adaptability is lower than of genotypic means.