Aim: Using a comprehensive floristic database (2616 species, 36,004 occurrence records from 128 unique localities), we model species turnover along the central region of the Atlantic Forest hotspot to (1) test whether local rivers, particularly the Rio Doce, are associated with marked biogeographical breaks, and (2) investigate how regional compositional changes correlate with geo-climatic variables. Location: The central region of the Atlantic Forest in eastern Brazil (12°–22° S latitude). Methods: We combine occurrence and geo-climatic data in a generalized dissimilarity model, obtaining a continuous prediction of species turnover across space and identifying 12 significant geo-climatic predictors of community composition. We use a two-step cluster analysis to classify the turnover map into major floristic regions based on the natural subgroups observed. We further divide each major floristic region into smaller sub-regions based on natural subgroups statistically identified by the two-step cluster analyses. Results: High levels of turnover in species composition occurred around latitudes 18°–19° S, c. 50–100 km north of the Rio Doce, and concurred with shifts in availability of both humidity and energy. We identified three major floristic regions in the central region of the Atlantic Forest, which we called Bahia Interior Forests, Bahia Coastal Forests, and the Krenák-Waitaká Forests – each of them divided into two to four subregions. Main conclusions: Our results suggest that local climatic conditions, not riverine barriers, drive biogeographical shifts in this region – a finding that supports studies of current and historical determinants of the composition of the Atlantic Forest biota. Floristic composition at higher elevations (> 600 m) is clearly distinct from those in lower elevations, likely as a result of physiological constraints imposed by cooler climates in the former. Floristic regions here identified from observed communities substantially improve the maps currently employed for conservation planning in a shrinking hotspot.