The method proceeds through two basic steps. In the first, a time separation threshold of 0.3 s is set up around each successive source time, which isolates candidates of other sources that might be linked in the same flash. In the second step, a range dependent distance threshold is applied to the candidate sources, which accounts not only for the physical distance between sources within a given flash, but also for the quadratic increase in radial position error and linear increase in tangential position error that are known to affect the source data. At ranges beyond about 50 km, the radial position uncertainty is the dominant factor in this spatial thresholding step. Accurate representation of flash events is found to be limited to ranges less than about 160 km from the network centroid, beyond which the source position error begins to compete with the spacing between convective cells. The optimum choice for the coefficient of quadratic increase in radial position error is problematic, but the degree of realism may be assessed by examining the range dependence of flash statistics such as flash duration. Preliminary comparisons with output from other flash algorithms shows our flash counts to be comparable or somewhat smaller. Nevertheless, the maximum single-storm cell flash rate measured by the algorithm so far is nearly 14 flashes per second, in a large tornadic supercell in Alabama on 10 November 2002.