Information about sample adequacy that represents soil chemical attributes distribution are fundamental for a better rationalization of the use of correctives and fertilizers. The objective was to evaluate the variability of these attributes and to size the minimum number of composite samples to represent the fertility of forest soils. The total area planted was 9,101ha, constituted of 265 commercial eucalypt stands. The 687 soil composite samples obtained were for chemical analysis. It was evaluated the performance of two exploratory analysis techniques and six sampling procedures. The attributes P, K, Ca, Mg and S presented higher coefficient of variation (>35%). In contrast, the distributions of Al, organic matter and, mainly, pH were the most homogeneous. The sample error was smaller as the amount of composite samples increased. The representative of all chemical attributes (sample error of 5%) was achieved with a minimum of 309 (one each 29ha, 1:29) and 295 (1:31) composite samples from sampling procedures simple casual and stratified by altitude class, respectively. Both procedures were promising for soil sampling, especially, when applying the boxplot for identification and removal of outliers.