BACKGROUND: Diagnostic tests enabling general practitioners (GPs) to differentiate rapidly between pneumonia and other lower respiratory tract infections (LRTIs) are needed to prevent increase of bacterial resistance by unjustified antibiotic prescribing. AIMS: To assess the diagnostic value of symptoms, signs, erythrocyte sedimentation rate (ESR), and C-reactive protein (CRP) for pneumonia; to derive a prediction rule for the presence of pneumonia; and to identify a low-risk group of patients who do not require antibiotic treatment. DESIGN OF STUDY: Cross-sectional. SETTING: Fifteen GP surgeries in the southern part of The Netherlands. METHOD: Twenty-five GPs recorded clinical information and diagnosis in 246 adult patients presenting with LRTI. Venous blood samples for CRP and ESR were taken and chest radiographs (reference standard) were made. Odds ratios, describing the relationships between discrete diagnostic variables and reference standard (pneumonia or no pneumonia) were calculated. Receiver operating characteristic analysis of ESR, CRP, and final models for pneumonia was performed. Prediction rules for pneumonia were derived from multiple logistic regression analysis. RESULTS: Dry cough, diarrhoea, and a recorded temperature of > or = 38 degrees C were independent and statistically significant predictors of pneumonia, whereas abnormal pulmonary auscultation and clinical diagnosis of pneumonia by the GPs were not. ESR and CRP had higher diagnostic odds ratios than any of the symptoms and signs. Adding CRP to the final 'symptoms and signs' model significantly increased the probability of correct diagnosis. Applying a prediction rule for low-risk patients, including a CRP of < 20, 80 of the 193 antibiotic prescriptions could have been prevented with a maximum risk of 2.5% of missing a pneumonia case. CONCLUSION: Most symptoms and signs traditionally associated with pneumonia are not predictive of pneumonia in general practice. The prediction rule for low-risk patients presented here, including a CRP of < 20, can considerably reduce unjustified antibiotic prescribing.