The aim of this study was to analyze the spatial pattern and sampling methods for Bertholletia excelsa. The inventory of exploration, by mapping Cartesian coordinates of all trees with dbh ≥ 20 cm, was carried out in an area of 1,000.31 ha. For the definition of the diametric structure, multivariate techniques were used: cluster and discriminant analysis. To analyze the spatial patterns, individuals were grouped into three levels: population, young individuals and adults; being the random deviation defined by the Ripley’s K function. The forest inventories were simulated with a simple random sampling, systematic and adaptive, taking into account plots of 2,500 m², with a sampling intensity of 15 % and an error range of 10 %. Comparisons between sampling methods were performed for the accuracy and precision. Four hundred and forty six trees were registered, of which 59 were young individuals and 387 were adults. A low number of trees in the early stages and a high number of individuals in the intermediate classes characterize the diametric structure. The spatial pattern of the adult population was aggregated, while the young individuals were randomly distributed. Systematic sampling is the best procedure to estimate the total number of individuals of B. excelsa; however, there is a need to investigate the effect of aggregation and size of the largest plots of 0.25 hectares on the estimates of adaptive cluster sampling.