16S and 23S rRNAs from Escherichia coli were used to study the relationship among a representative collection of strains of Erwinia chrysanthemi differing in their original host and geographical origin. Phenetic analysis of restriction fragment length polymorphisms allowed the distribution of the studied strains into seven clusters. These clusters were similar to those obtained by cladistic methods and appeared to correlate well with the established pathovars and biovars but to a lesser extent with geographical distribution. Except for two groups of strains defined as tropical and temperate isolates (clusters 3 and 4, respectively), our clustering correlated well with botanical classifications of host plants. However, the rRNA groupings were shown to be more discriminative than biovar analysis. To assess the relationship between rRNA clusters and pathogenicity, 12 representative strains from different clusters were tested for pathogenicity on different plants. The two typical symptoms, maceration and wilting, were observed for these strains. The occurrence of the tobacco hypersensitivity reaction for a subset of these strains is discussed in light of recent results concerning the presence of an hrp gene. Considering symptom expression only, rather than the capacity for plant infection, strains from the same cluster were shown to induce similar symptoms in test plants. Thus, since host specificity is still quite controversial, rRNA patterns may constitute a useful tool in taxonomic and epidemiological studies of Erwinia chrysanthemi species.