Pneumonia severity assessment systems such as the pneumonia severity index (PSI) and CURB-65 were designed to direct appropriate site of care based on 30-d mortality. Increasingly they are being used to guide empirical antibiotic therapy and also possibly to detect patients who will require admission to the intensive care unit (ICU). We retrospectively reviewed the records of all patients admitted to our institution with confirmed community acquired pneumonia (CAP) for the 12 months from January 2002. 408 episodes were studied with an overall 30-d mortality of 15.4% and ICU admission of 10.5%. PSI classes IV/V were significantly better than CURB-65 score > or = 3 for predicting patients who died within 30 d (94% vs 62%; p < 0.001), and those that needed ICU (86% vs 61%; p = 0.01). In addition, for the patients identified as 'low risk' by PSI (classes I/II), there was only 1 death and 1 admission to an ICU compared to 8 deaths and 7 ICU admissions with CURB-65 scores of 0-1. Although easier to use, CURB-65 is neither sensitive nor specific for predicting mortality in CAP patients. Neither rule was sufficiently accurate for predicting need for an ICU, even when patients with 'not for resuscitation' orders were excluded.