The Rhizobium strain specificity and effectiveness in nitrogen fixation of 336 accessions of the genus Stylosanthes were determined in glasshouse screening trials. In preliminary screening, against a wide-spectrum strain CB756, 224 accessions were effective (> 50% dry weight of nitrogen control). Sixty-two of these were evaluated subsequently against a spectrum of 22 strains of Rhizobium selected on the basis of host and geographic origins. Thirty-three were effective with 10 or more strains. Seventy-two accessions that were ineffectively nodulated in preliminary tests were also screened against the 22 strains. Fifty-four of these were effective with one or more strains of Rhizobium. A minimum spanning tree (MST) analysis was used to examine the interrelationships between, and the distribution of, accessions according to their effectiveness response (ER) to the 22 strains of Rhizobium. Accessions were classified into six groups, also according to their effectiveness response, by the program MULTBET. Groups I–IV were nodulated effectively by a large number of strains (mean 9.4) and groups V and VI by a limited number of strains (mean 1.3). Significant and wide-ranging Stylosanthes accession x strain of Rhizobium interactions for effectiveness in nitrogen fixation were observed, particularly in S. guianensis and S. hamata. Various reclassifications were made by MULTBET on a reduced number of strains which were selected on the basis of discriminatory powers between groups (eident values), contribution to the classification (Cramér measures) and intuitively on the basis of level of effectiveness specificity and soil pH adaption. Classifications based on Cramér measures gave least (9%) non-conforming accessions, but intuitively selected strains the widest range of effective associations. The majority of non-conforming accessions were found in effectiveness response groups ER-III and ER-IV and were mostly S. guianensis from M–A groups 2 and 14 andS. hamata M–A 28. For future screening programs strains CB 82, CB 159, CB 530, CB 756, CB 1408 and CB 2126 will be used as diagnostic strains for classification of accessions by effectiveness response, and strains CB 82, CB 1650 and CB 2126 as 'key strains' for provision of inocula for experimental purposes.Classification of strains of Rhizobium into six groups was obtained by a MULTCLAS analysis of the data matrix. A principal coordinates analysis suggested distinct clustering of strains on the basis of host specificity for effective nodulation. These corresponded very closely to the six classificatory groups.