Analysis of the biogeographic distributions of bacteria has been limited by potential biases inherent in the isolations required for classical taxonomy and by the time required for phylogenetic analyses. We have attempted to circumvent both of these limitations by using denaturing gradient gel electrophoresis (DGGE) to resolve the products of polymerase chain reaction (PCR) amplifications of mixed template DNA isolated from microbial communities. DGGE separates DNA fragments based on their denaturation characteristics, which vary with the nucleotide sequence of the fragment. The banding patterns in the electropherograms were then subjected to similarity analysis using pattern matching and band comparison software. Replication experiments tested the robustness of band patterns within and between gels. Samples were collected from the Central Arctic Ocean basin during April of 1995 on the SCICEX 95 cruise of the USS Cavalla. One hundred samples collected from a depth of 59 m are the focus of this biogeographical analysis. The band identification algorithm of the software identified between 12 and 30 bands (operational taxonomic units, OTUs) per sample (mean: 21.5) with minimal editing. This number approximately doubled with more extensive editing. Four OTUs seemed to be common to most samples. The samples grouped into five major clusters with similarities greater than approximately 80%. Twenty nine samples in one of these clusters were in two branches with internal similarities greater than approximately 90%. These samples had relatively nondescript banding patterns (numerous bands with roughly equal intensity). Another cluster contained 15 samples with distinctive banding patterns dominated by one or two intense bands. These samples were collected in the same general area of the Arctic Ocean (Canada Basin) and may reflect a community response to local environmental conditions.