We examined distribution and abundance of small, medium, and large American eels Anguilla rostrata in Virginia streams by comparing observed with expected null-model patterns. At large spatial scales (across drainages or physiographies), densities of small and medium American eels decreased with distance from the ocean, and densities were not strongly or consistently related to local habitat features (e.g., habitat type, abundance of cover, water temperature, density of predators). A mathematical function, analogous to diffusion of particles from a concentrated source, accounted for up to 85% of the variance in densities of small- or medium-sized eels at statewide, physiographic, or river-drainage scales. At smaller spatial scales (e.g., within a drainage in a particular physiography), we found few relations between American eel density and habitat features. Exceptions to this pattern occurred at Coastal Plain sites of the James River drainage, where American eel densities were exceptionally high. Such small- scale, nonrandom patterns may reflect local density-dependent processes that only rarely affect American eel distribution and abundance more than do ubiquitous, large-scale random processes. Parsimonious, large-scale models (e.g., diffusion models) may predict the distribution and abundance of American eels (and other widespread or generalist stream fishes) more reliably than do more typical, small-scale habitat models (e.g. habitat suitability models).