Earthquakes occur in clusters, which classically are described as foreshock–mainshock–aftershock sequences or swarms. In this paper, every earthquake in a seismicity catalog is assigned to a cluster if it is separated from at least one other event in the cluster by less than Δt in time and less than Δr in space. The minimum cluster size is one earthquake. For catalogs that are complete to small magnitudes, this approach is successful in capturing the full spatial extent of an extensive cluster even for Δr much smaller than the actual cluster dimension. The declustered catalogs are much closer to Poissonian distribution than the originals. This was applied to seismicity catalogs for Japan, Southern California, and Nevada. Cluster sizes measured by the number of earthquakes in the cluster exhibit an approximate power‐law frequency distribution. An upper bound to cluster durations is proportional to K0.5, where K is the number of earthquakes in the cluster. This paper demonstrates an analytical approach suitable for selecting values of Δt and Δr that are appropriate for the earthquake catalog.