Gas chromatographic fingerprints of eighty-three strains of Streptococcus were analysed by computer. Seven measures of association were compared for their ability to identify strains. The most effective measure was the Stack coefficient which correctly identified 68% of strains, mostly of oral origin. Clustering of strains was carried out by median, average, single-linkage, furthest neighbour, and centroid linkage, Andrew's plots, and Minimum Spanning Trees. Of the clustering methods, centroid linkage produced the most inclusive and compact clusters. Clusters of strains of S. mitis, S. mutans, S. salivarius and S. sanguis showed varying degrees of heterogeneity; while, S. milleri was comprised of three chemotypes corresponding to oral isolates, biochemically atypical vaginal isolates, and biochemically typical strains from other sources.