To assess community-acquired pneumonia severity scores from two perspectives: (i) prediction of ICU admission or mortality; and (ii) utility of low scores for prediction of discharge within 48 h, potentially indicating suitability for short-stay unit admission. Patients with community-acquired pneumonia were identified from a prospective database of emergency patients admitted with infection. Pneumonia severity index (PSI), CURB-65, CORB, CURXO, SMARTCOP scores and the Infectious Diseases Society of America/American Thoracic Society (IDSA/ATS) minor criteria were calculated. Diagnostic accuracy statistics (sensitivity, specificity, predictive values, likelihood ratios and area under receiver operating characteristic curves [AUROC]) were determined for both end-points. Of 618 patients admitted with community-acquired pneumonia judged eligible for invasive therapies, 75 (12.1%) were admitted to ICU or deceased at 30 days, and 87 (14.1%) were discharged within 48 h. All scores effectively stratified patients into categories of risk. For prediction of severe pneumonia, SMARTCOP, CURXO and IDSA/ATS discriminated well (AUROC 0.84-0.87). SMARTCOP and CURXO showed optimal sensitivity (85% [95% confidence interval (CI) 75-92]), while specificity was highest for CORB and CURB-65 (93% and 94%, respectively). Using lowest risk categories for prediction of discharge within 48 h, only SMARTCOP and CURXO showed specificity >80%. PSI demonstrated highest positive predictive value (31% [95% CI 24-39]) and AUROC (0.74 [95% CI 0.69-0.79]). Community-acquired pneumonia severity scores had different strengths; SMARTCOP and CURXO were sensitive with potential to rule out severe disease, while the high specificity of CORB and CURB-65 facilitated identification of patients at high risk of requirement for ICU. Low severity scores were not useful to identify patients suitable for admission to short-stay units.