A total of 254 strains of the genera Nocardia, Rhodococcus, Amycolata, Amycolatopsis, Gordona and Pseudoamycolata were compared using numerical taxonomic techniques based on 273 physiological characters with the aid of miniaturized tests. Clustering was achieved using the unweighted pair group method with arithmetic averages (UPGMA) and the simple-matching coefficient (S_SM) as the measure for similarity. Test error and overlap between the phena were within acceptable limits. Cluster groups were defined at the 87.4 to 91.3% levels (S_SM). A total of 25 clusters were obtained in the S_SM/UPGMA analysis beside 16 single member clusters, which in most cases were marker strains of different species. For the 25 clusters, containing two or more strains, a matrix comprising frequencies for positive results of 35 tests was constructed. The minimum number of diagnostic characters was selected by computer programs (CHARSEP, DIACHAR, MOSTTYP). The final matrix consisting of 35 tests versus 25 phena was theoretically evaluated using a computer program (MATIDEN) and out of 238 strains, a total of 157 strains (65.96%) were correctly identified with a Willcox probability>0.999, further 18 strains (7.56%) achieved Willcox probabilities above 0.900 and 2 strains (0.84%) were identified correctly with a Willcox probability above 0.800. Correct identification results were obtained for a total of 56 strains (23.52%), but with low Willcox probabilities ranging between 0.500 and 0.800. A total of 5 strains (2.10%) could not be assigned to the correct phenon. In a subsequent practical evaluation of the matrix 32 of 40 tested strains (80.0%) were correctly identified with Willcox probabilities above 0.900.