Cross-validation is used to select an optimal subset from the r-best (r 10) subset models identified by the branch-and-bound regression algorithm given by Furnival and Wilson (1974) with selection criterion Mallows' Op. A statistic, predicted-Cp (a function of PRESS analogous to Mallows' Op statistic), is proposed as a criterion function in comparing the r-best subset models. The optimal subset is that which has the minimum predicted-Cp. Two examples are presented.