This study considers the selection of the best few ofafinite population of varieties whose unknown true yields are normally distributed. The selection scheme consists of a series of yield trials each utilizing a specified proportion of the fixed total resources. At each stage certain varieties are rejected; the rest are passed on to the next stage. Increased yields are found to result when rejection is based on information from all previous stages and on external cut-off points rather than a variety's relative position within a cohort.