This study was designed to investigate the value of biologic evidence of inflammation for the diagnosis of giant cell arteritis. Experienced physicians were asked to evaluate five pairs of medical records based on real cases. In each pair, one case lacked biologic evidence of inflammation. This study offered the opportunity to explore the feasibility of a simplified Bayes model. A blind evaluation obtained by showing the paired case-reports with similar evidence of inflammation in both cases of each pair to 14 specialty physicians yielded a likelihood of diagnosis of +/- 20%. Analysis of the 46 responses to the study demonstrated, despite wide variations, a significantly greater likelihood of diagnosis in the cases with evidence of inflammation. Nevertheless, 17% to 36% of physicians--according to the case-report--ascribed virtually no importance to the ESR. Most of the physicians considered temporal artery biopsy was warranted when the likelihood of diagnosis was greater than 25%. Emergency corticosteroid therapy while awaiting the histologic results was approved by most responders when the likelihood of diagnosis was greater than 65%. The "pre-test" likelihood, calculated assuming that sensitivity and specificity of the ESR are 0.99 and 0.50, respectively, ranged from 0.89 to 0.98 for the case-reports with no evidence of inflammation and from 0.16 to 0.59 for the case-reports with evidence of inflammation In theory, the figures for the two types of case-report should not differ by more than 20%. Use of a low value for specificity (0.05) would improve the fit of values in cases without evidence of inflammation but would increase discrepancies in the other cases.(ABSTRACT TRUNCATED AT 250 WORDS)