Areas of California and Nevada that are particularly earthquake prone are identified with a computer by applying a pattern-recognition algorithm to standard geological data. The algorithm, which is designed for varied application, defines suites of characteristic traits and successfully discriminates earthquakes characteristic of the San Andreas fault system from those characteristic of the western Basin and Range province. It also pinpoints areas in California and Nevada that are unlikely to be the epicenters of strong earthquakes. Control experiments to test the predictive ability of this technique have shown imperfect but positive results. Predictions of future earthquake epicenters are made. Features characteristic of earthquake-prone areas emerge that appear physically meaningful in terms of large-scale geology. Some of these characteristics were not obvious before.