Let there be given k(2 1) independent and normally distributed populations nl, ..., nk where ni has unknown mean pi and unknown variance oi2 (i=l, ..., k). A single-stage sampling procedure is developed for constructing confidence intervals for the largest of the means. The method is to split up the single sample into two portions, the first consisting of n-1 observations for initial estimation and the second consisting of the remaining one for use as a spare in the final estimation. One-sided, two-sided, and asymptotically optimal confidence interval are considered. A numerical example to apply this procedure is given.