Abstract: Anew model selection method named orthogonalization recurrence selection (ORS) method is proposed.By use of this method the model with higher predictive ability can be ob-tainded or the lower PRESSstatistic values of the model can be achieved.The comparison is made among forward selection method in the stepwise regression and PLSas well as ORSmethods through the example of QSAR(quantitative structure activity relationship).

Key words: Model selection, PRESS criterion, Stepwise regression, PLS regression, Gram-Schmidt ortllogonalization