Brook trout Salvelinus fontinalis are native fish in decline owing to environmental changes. Predictions of their potential distribution and a better understanding of their relationship to habitat conditions would enhance the management and conservation of this valuable species. We used over 7,800 brook trout observations throughout New York State and georeferenced, multiscale landscape condition data to develop four regionally specific artificial neural network models to predict brook trout abundance in rivers and streams. Land cover data provided a general signature of human activity, but other habitat variables were resistant to anthropogenic changes (i.e., changing on a geological time scale). The resulting models predict the potential for any stream to support brook trout. The models were validated by holding 20% of the data out as a test set and by comparison with additional field collections from a variety of habitat types. The models performed well, explaining more than 90% of data variability. Errors were often associated with small spatial displacements of predicted values. When compared with the additional field collections (39 sites), 92% of the predictions were off by only a single class from the field-observed abundances. Among “least-disturbed” field collection sites, all predictions were correct or off by a single abundance class, except for one where brown trout Salmo trutta were present. Other degrading factors were evident at most sites where brook trout were absent or less abundant than predicted. The most important habitat variables included landscape slope, stream and drainage network sizes, water temperature, and extent of forest cover. Predicted brook trout abundances were applied to all New York streams, providing a synoptic map of the distribution of brook trout habitat potential. These fish models set benchmarks of best potential for streams to support brook trout under broad-scale human influences and can assist with planning and identification of protection or rehabilitation sites.