One hundred and eight strains of Agrobacterium isolated from soil and from affected plants of 13 species, together with type and representative strains, were subjected to cluster analysis using 28 phenotypic characters out of the 36 examined. Data were analyzed by applying the unweighted average pair-group method (UPGMA) on the simple matching coefficient (S_SM). At 80% S_SM, 5 clusters were recognized, 3 of which corresponded to biovars 1, 2 and 3. These biovars were each grouped into a tight cluster, indicating that each of them is phenotypically homogeneous. Two additional clusters comprised (a) 3 overseas strains (IFO 13261 and 13260 of A. rubi, and NCPPB 1650 of A. tumefaciens), and (b) 6 Japanese strains of A. tumefaciens isolated from cherry and kiwifruit. The former cluster neighbored biovar 3 and the latter biovar 2, respectively. All strains in each biovar gave uniform reactions for the following 15 characters, which are therefore considered to be useful for the differentiation and the identification of the biovars: production of 3-ketolactose; esculin hydrolysis in Sneath's medium; arbutin hydrolysis; arginine dihydrolase in Thornley's medium; growth factor requirement; litmus milk reaction; pellicle in ferric ammonium citrate solution; growth at 35°C, and on New and Kerr medium; utilization of citrate, and L-tyrosine; oxidase by growth on PPGA; acid from dulcitol, and α-methyl-D-glucoside; and alkali from L-tartrate.