CONTEXT: Although observational studies suggest that clinical pathways may decrease costs and improve quality in hospitalized patients with community-acquired pneumonia, inferences from these studies are limited by potential selection bias and inadequate case-mix adjustment. OBJECTIVE: To compare the assessment of a clinical pathway for community-acquired pneumonia with and without adjusting for patient characteristics and disease severity. DESIGN: Retrospective cohort study. PATIENTS AND SETTING: Consecutive series of adult patients admitted with clinical diagnosis of community-acquired pneumonia, treated with either a clinical pathway (which included guidelines for antibiotics, tests, and ancillary care) or usual care. MAIN OUTCOME MEASURES: Total hospital charges, length of stay, clinical deterioration (requiring mechanical ventilation or intensive care unit transfer), and in-hospital mortality. We used multiple linear and logistic regression to adjust for patient case mix. RESULTS: Compared with patients receiving usual care (n = 275), patients in the pathway group (n = 97) were more likely to be treated by family physicians than specialists and had lower pneumonia severity scores. In the unadjusted analysis, total hospital charges were lower among pathway patients ($2456; 95% CI, $175 to $4737; P = 0.04); in the adjusted analysis, the difference in total charges was smaller (average reduction $1807; CI, $4164 lower to $549 higher; P = 0.13). In the unadjusted analysis, length of stay was lower among pathway patients (1.8 days lower; CI, 3.9 lower to 0.4 higher; P = 0.12); in the adjusted analysis, the difference in length of stay was smaller (0.9 days lower; CI, 3.2 lower to 1.3 higher; P = 0.4). Although unadjusted analysis showed significantly lower in-hospital mortality in pathway patients, this difference was not confirmed in the adjusted analysis. CONCLUSIONS: Clinical pathways may reduce costs and improve quality of care in community-acquired pneumonia. In nonrandomized studies, however, selection bias and case-mix differences may explain some of the apparent effectiveness.