The pattern-recognition technique of Bongard and others (1966), Gelfand and others (1976), and Briggs and others (1977) was modified and applied in Southeastern United States to try to detect combinations or interactions of geologic, geophysical, and topographic parameters characteristic of areas with moderate and large earthquakes. The modification involved replacing the operator-defined criteria for significance of association of parameters and earthquakes with statistical estimates of significance. This was done in order to decrease the probability of random associations and to increase the flexibility of the pattern-recognition technique. The binomial distribution was used to determine the probability that each particular association could have occurred by chance. Bonferroni's inequality was invoked to determine a conservative significance level for each of the binomial tests. However, without being able to quantify the dependence among the associations, it is not possible to statistically evaluate the group of binomial tests as a whole. Preliminary results identify four independent associations characterizing spatially independent earthquake epicenters, or "dangerous" points, and seven independent associations characterizing the intervening "safe" points. When all points were reclassified using these traits, all epicenters were correctly classified as dangerous. In addition, three "safe" points were reclassified as dangerous and appear to be potential sites for future earthquakes. However, it is not possible to determine the degree of dangerousness of the points using this reclassification scheme. Two dangerous zones emerge in the northeast and southwest portions of the study area. Epicenters outside of these zones, including those of the large historic earthquakes (MMI > VIII) in Giles County, Virginia, and Charleston, South Carolina, have not been well characterized by the significant associations and may have fundamentally different causes.
