Abstract Aim In eastern Nepal, forests occupy an elevation gradient of 4000 m with bioclimatic zones from near tropical to alpine. Understorey plants and trees were censused to measure species density and identify patterns of ecological change. By sampling in a manner robust against spurious mid-domain effects, I aim to identify biologically valid controls on species density. Location The study area consists of land below 4250 m elevation between 27.1 and 27.8° N latitude, 86.5 and 88.0° E longitude on the southern slopes of the Himalaya Range in eastern Nepal. Sampling sites are limited to intact, natural forest with relatively little human impact. Methods Team members counted species of understorey plants and trees ≥ 10 cm d.b.h. in 0.04 ha plots throughout the study area. In addition, basal area, leafing phenology and species composition were determined for the trees in each plot. Estimates of regional species density were compiled for successive 250 m elevation bands from 250 to 4250 m elevation. Species density trends were identified and compared with the expectations of O'Brien's [Journal of Biogeography25 (1998) 379–398] climate-based water–energy dynamics model. Results Stand basal area, tree leafing phenology and taxonomic composition (angiosperm vs. gymnosperm) show non-random change with elevation. Understorey plant and tree species density both have a humped, unimodal trend with more species near the bottom of the gradient and fewest at the top. These trends are consistent with expected effects of the climatically active water and energy variables. After curve-fitting, significant spatial structure in the residuals suggests that tree communities within the 1750–2250 m elevation range do not realize their climatic potential species richness. Main conclusions Neither mid-domain effects nor biologically valid boundary effects like dispersal limitation explain the plant species density trends observed. Trends do fit a model in which species density is controlled by the same ‘active’ climatic variables that predict species richness on continental scales. Patterns of leafing phenology on the elevation gradient provide further support for the hypothesis of environmental control of species density. The productivity–diversity linkage that exists on continental scales may also apply on the smaller scale of a Himalayan elevation gradient. Human activity and possible competitive exclusion by Castanopsis tribuloides are the two best explanations for the observed decline in tree species density at 1750–2250 m elevation. Burning, lopping for fodder and livestock grazing might account for the decline, but this study does not assess the relative importance of these activities. The elevation richest for understorey plant and tree species (500–1500 m) also has the most severe reduction in forest cover. Local farmers deserve credit for sustaining plant biodiversity in forest enclaves, but further loss of forest at these elevations should be discouraged.