The species–area relationship (SAR) on islands describes how the numbers of species increase with increasing island size (or island-like habitat, such as lakes). Despite its conceptual importance, there is considerable uncertainty surrounding its shape in freshwater lakes, as well as the potential mechanisms that underlie the SAR. Here, we used standardized sampling data of benthic macroinvertebrates from 81 shallow lakes in the Yangtze-Huai floodplain of China to evaluate its shape and disentangle several mechanisms (e.g., passive sampling, colonization-extinction dynamics and heterogeneity) underlying the SAR. At the whole-lake level, we found an increase in the total species richness with increasing lake area, as well as an increase in rarefied richness controlling for sampling effort. However, within single samples, diversity was negatively related to lake area. This scale-dependence is because within-lake β-diversity increased with lake area, suggesting that heterogeneity overrides local processes to generate the positive SAR. These patterns were only evident for measures of diversity that equally weigh common and rare species (i.e., species richness), and disappeared when diversity measures that weigh common species more heavily were used. This suggests that the influence of heterogeneity on the SAR was largely via its influence on rarer species and their turnover in larger lakes. Overall, our result that heterogeneity was the primary driver of the positive SAR in this system provides an important baseline for making predictions about biodiversity changes with ongoing habitat loss.