The errors of 454 cereal experiments organized by the National Institute for Agricultural Botany between 1956 and 1960 are examined in relation to block size. All the experiments used long, narrow plots and were harvested by combine. The most important conclusion is that the gain in efficiency by using small blocks in cereal experiments is as great now as it was before combine-harvesting and long, narrow plots were introduced. An empirical rule is that the variance per plot is roughly proportional to n½, where n is the number of plots per block.