Cross-validation is a popular re-sampling technique used for assessment of statistical models and selection amongst competing regression models. Salahuddin and Hawkes (1991) used cross validation in stepwise regression to select appropriate cutoff values, by putting F-to-enter equals F-to-remove, which minimizes PRESS. However, several other researchers have suggested making F-to-remove smaller than F-to-enter (for example see, Graybill and Iyer (1994); Miller (1996). In this study, we have used cross-validation technique to choose unequal cutoff values (by taking F-to-enter smaller than F-toremove) in stepwise regression, which are then used to determine predictor variables in the final regression model. The procedure selects the optimum unequal cutoff values which produces the minimum stable PRESS. Furthermore, based on the application of cross-validation to optimize unequal cutoff values in steptwise regression procedure, we detect true outliers and influential observations. Some examples are used for demonstration purpose. The study found that the proposed technique is working well.