Physical factors that may account for regional patterns of plant species diversity remain controversial. We aim to determine the relationship of tree species diversity to environmental factors identifiable at regional scale in the northern Neotropics. We use a high-resolution data set based on herbarium collections of all native tree species known to occur in the highly diverse and physiographically heterogeneous Mexican state of Chiapas. We analyzed 114 grid cells (5 min latitude×5 min longitude each) with 40 or more vouchers. We obtained from maps (scale 1:250 000) data on temperature, rainfall, elevation, and soils, and calculated for each grid cell mean actual evapotranspiration (AET), its ratio during the rainy and dry seasons (RET), average fertility/quality of soils (SFQ), and elevation (coarse-scale topography) variance (SDE). These variables were largely independent of each other, and were entered in multiple regression models to predict species diversity assessed with Simpson's index of concentration. A model that accounted for 41.4% of the total variance in tree diversity showed positive effects of AET and seasonality (RET), whereas SFQ had a negative effect. A curvilinear model described well the relationship between tree diversity and AET (R2=0.45), and an intermediate maximum was detected. The data pattern suggested an asymptotic relationship as well, which was confirmed with a two-part regression. Regression quantiles provided better estimates of the effect of SFQ with the upper envelope of the data (0.85–0.90 quantiles). Minimum diversity at intermediate rainfall values hints at a bimodal model of tree diversity along a rainfall gradient, in opposition to the frequent contention of a positive linear relationship. We suggest that broad-scale climatic gradients interact with intraregional landscape-level influences, thus leading to the observed nonlinear responses of tree diversity to environmental predictors.