The preliminary American Rheumatism Association (ARA) criteria for classification of systemic lupus erythematosus (SLE) were evaluated for sensitivity and specificity in 87 patients with SLE and 73 without SLE from a rheumatology population. Using these data, Bayes' theorem was employed to weight the criteria so that a scoring system could be developed, allowing more accurate use of the criteria in diagnosis. Combinations of serologic test results were evaluated in a similar manner. Comparison of weighted scores with criteria counts showed greater sensitivity and specificity of the former whether the 1971 or modified 1982 ARA criteria were used. Weighted criteria may be more useful in clearly defining patient populations for studies of SLE.