Grain (size of sampling units) affects the spatial resolution at which ecological patterns can be observed and analysed, and potentially has an important effect on the results of broad‐scale studies on diversity gradients. Here we examine the effect of grain on patterns of species richness and turnover in lowland rainforests of western Amazonia (Peru and Colombia). We inventoried pteridophytes (ferns and lycophytes), melastomes (Melastomataceae) and palms (Arecaceae) in four line transects of 22–29 km length. Different grains were obtained by aggregating original 100‐m‐long sampling units into larger segments up to 19.2 km long. With any given grain and plant group, local species richness varied considerably both within and among transects, and a transect segment that was species‐rich with one grain could be relatively species‐poor with another. Which transect had the highest vs lowest mean species richness per sampling unit (α richness) differed among plant groups. It also varied to some degree with grain, as transects differed in how rapidly local species richness increased with increasing grain. Patterns of species turnover were more consistently correlated among plant groups than patterns of species richness were, and NMDS ordinations were rather similar with all grains and plant groups. Floristic heterogeneity within the Amazonian terra firme rainforest seems to contain a general compositional pattern that is sufficiently robust to be detectable with various sampling schemes, but patterns of species richness appear more case‐specific. Therefore, using one plant group as an indicator for patterns in other plant groups can be expected to work better for species turnover than for species richness.