Uniformity trial with sugar cane was carried out in the Usina da Barra, at Lençóis Paulista, State of São Paulo, Brazil. Based on 1,512 yield data of sugar cane harvested in areas of 3.0m² (basic unit) or 1.5m by 2.0m, 55 different types of plots were simulated. The Smith's soil heterogeneity index, b, was estimated. Its values varied from 0.2643 to 0.6000, depending on the method employed or on the way of grouping blocks, plots and subplots. Using b = 0.6000 in a formula given by Smith, x= (b/(1-b)(K1/K2),an area of 6.0m² would give more information at lower cost. By the maximum curvature method the greatest reduction in the value of coefficient of variation occurs when the plot size varied from 6.0 to 12.0m². Presently, most plots used in sugar cane experiments range from 26.0 to 60.0m²; they are too large in relation to the best size obtained, that is, around 12m². There are practical advantages in using smaller plots and larger number of replications of treatments in an experiment, because the mean variance is inverselly proportional to the number of replications. Reducing the mean variances implies greater chance of detecting significant differences between treatment means. For homogeneous soils, smaller plots (12 to 36m²) may be used efficiently. In heterogeneous soils the plot size has little effect in the magnitude of the index d. The influence of the plot length, in reducing the coefficient of variation, was 2.6 times greater than the plot width. Therefore, in order to separate small differences between treatments, it is recommended the use of plots of one line 12m long with 12 replications per treatment or two lines 8m long with 8 replications. The plots considered in this paper are without guard rows.
sugar cane; plot; plot size; experimental precision