The usefulness of the sputum Gram's stain is controversial. This meta-analysis was designed to evaluate the sensitivity and specificity of the sputum Gram's stain in community-acquired pneumococcal pneumonia. Using a predetermined protocol, articles were discovered through a MEDLINE search (1966 to 1993) and the examination of bibliographies and were graded for quality by three blinded reviewers. Information on the reference standard, blinding, stain interpreter, control for antibiotic use, and definition of a positive test was collected. We found 12 articles containing 17 test characteristics to evaluate. The number of patients in each study ranged from 16 to 404. Sputum culture was the most common reference standard (10 of 17 estimations). Sensitivity ranged from 15% to 100% and specificity from 11% to 100%. Test characteristics varied markedly among studies and appeared related partly to the test interpreter. The sputum Gram's stain may yield misleading results in community-acquired pneumonia, as its sensitivity and specificity vary substantially in different settings. A practitioner electing to use the study should be well trained and use a specific definition for a positive test.