An analysis was made of the floristic composition of 98 areas of cerrado and Amazonian savanna, encompassing most of the area of such vegetation in Brazil. A total of 534 species of trees and large shrubs were recorded for these areas, of which 158 (30%) occurred at a single site only. Such unicates and taxa without determinations to specific level were excluded from the study since they provide no basis for comparison. The data were analysed by three techniques of multivariate analysis: (a) a divisive hierarchical classification by Two-way Indicator Species Analysis (TWINSPAN). (b) an agglomerative hierarchical classification by UPGMA (Unweighted Pair-Groups Method using Arithmetic Averages) using the Sørensen Coefficient of Community (CC) as a measure of similarity, and (c) an ordination by Detrended Correspondence Analysis (DCA). The results from all three methods showed great similarity, demonstrating a strong geographic pattern in the distribution of the flora of the cerrado biome and allowing the recognition of southern (Sâo Paulo and S Minas Gerais), southeastern (largely Minas Gerais), central (Federal District, Goiás and parts of Minas Gerais), central-western (largely Mato Grosso, Goiás and Mato Grosso do Sul) and northern groups (principally Maranhão, Tocantins and Pará), as well as a disjunct group of Amazonian savannas. Soil type (mesotrophic or dystrophic) is an important factor in determining floristic composition. The study demonstrated that cerrado vegetation is extremely heterogeneous: none of the 534 species occurred at all sites and only 28 species were present at 50% or more.