The 339 cases of systemic lupus erythematosus (SLE) and controls used in 1982 to develop the current ARA rules for classifying this disease were further analyzed using recursive partitioning methodology. The method identified 9 subgroups of cases and controls. It allowed correct classification of 97% of cases and 95% of controls while using only 6 criteria instead of 11. Antibodies to DNA were found to be the overall best discriminator. The largest subgroup (116 cases/4 controls) was made up of patients with both anti-DNA and fluorescent antinuclear antibody (FANA). Presence of malar rash was helpful in distinguishing SLE cases among patients who were anti-DNA negative. Low and high FANA values, in combination with other variables, helped to identify 3 SLE subgroups. Unlike the 1982 criteria, which did not include serum complement, the classification tree also identified subgroups of patients in which complement was either the best or the 2nd best discriminator. Recursive partitioning methodology thus gives important insights into how the criteria interact and the relative importance of specific variables for classification purposes.