Yield testing in early generations of wheat breeding programmes is limited by large numbers of lines and small amounts of seed. This paper compares yield testing using conventional row plots, hill plots and small single-ear plots. The effects of smaller plot size are compensated for by increasing the number of replications when moving from row plots, through hill plots to ear plots. Statistics derived from standard ANOVA's revealed small ear plots to be as useful as larger hill or conventional row plots in assessing the yield potential of lines. In spite of rather large genetic as well as environmental coefficients of variation, ear plots had a lower intra-class correlation coefficient than hill plots or row plots while the repeatability value of ear plots was higher than that of hill or row plots. Spearman rank correlation coefficients showed a highly significant correspondence in the ranking of trial entries in the three systems tested.