The species–area relationship (SAR), which predicts species richness scales positively with increasing area, is among the longest-recognized trends in ecology. Despite its ubiquity, our knowledge of the precise shapes and underlying mechanisms driving SARs remains incomplete, particularly in freshwater ecosystems. I tested theoretical predictions for SAR shapes and best-fit models and evaluated SAR scale dependency across the nested sampling scales of sites (finer scale) and hydrological units (HUs; intermediate scale), while using stream fish communities from the Alabama aquatic biodiversity ‘hotspot’ as my model system. Using data from 599 standardized fish surveys, I conducted model selection and multimodel averaging on 20 SAR models (e.g., logarithmic, power, logistic) and I examined the influence of three main hypotheses influencing SARs—passive sampling effects, disproportionate effects, and heterogeneity. I also used multiple linear regressions to test environmental drivers of SARs. Despite model uncertainty, my results supported scale dependency of stream fish SARs and their best-fit models; however, more flexible nonlinear models generally outperformed other models, and the Monod (i.e., Michaelis–Menten) model provided a plausible fit across scales. Despite partial sampling effects, the null “passive sampling hypothesis” was rejected in favor of disproportionate effects of increasing area, mediated most prominently through additions of common species, rather than rarer species. Multiple regressions revealed that a greater percentage of variation in freshwater fish species diversity at the sampled scales is explained by area than latitude or other environmental factors. Understanding the mechanisms driving SAR results has important implications for conservation and management, including aiding reserve design as well as development of predictions for biodiversity change in response to habitat loss or fragmentation.