Our aims were to quantify and map the plant ecoregions of the Atlantic Forest, a biodiversity hotspot that covers ca 150 million ha in eastern South America. We used a data set on the distribution of 4378 shrub and tree species across 711 localities. Plant ecoregions were identified using analyses of species turnover for both species occurrences and relative abundances. We interpolated NMDS axes of compositional variation over the entire the Atlantic Forest extent, and then classified the compositional dissimilarity according to the number of biogeographical ecoregions previously identified through K-means analyses. We assessed the ability of environmental, historical vegetation stability and the current human footprint to explain the occurrence of the identified ecoregions through multinomial logistic regression models. We identified 21 spatially cohesive occurrence and 14 abundance ecoregions. Aridity, soil and historical biome stability were retained in the best model explaining both occurrence and abundance ecoregions. Broad compositional zones were identified through UPGMA cluster analysis of ecoregions, and formed north and south compositional blocks. Our work confirms the existence of a broad north-south divide within the Atlantic Forest, previously suggested based on climatic and amphibious data. Differences between the occurrence and abundance maps suggest the location of transition zones to neighbouring domains and endemism centres. Due to the aggregate nature of our analyses, site-level disturbance degree was not considered, implying that human impacts could be broader then we could detect. There was limited overlap between our results and previous Atlantic Forest regionalization efforts, indicating that multi-taxa, physiognomic and environmental regionalization schemes based on expert opinion or vegetation maps are poor proxies for compositional ecoregions.