Four hundred and fifty-three adults in 25 British hospitals entered a prospective study of community-acquired pneumonia. A microbiological diagnosis was established in 67 per cent; Streptococcus pneumoniae (34 per cent). Mycoplasma pneumoniae (18 per cent) and Influenza A virus (7 per cent) were the commonest microorganisms. Our observations support the view that most of those in the microbiologically negative group (33 per cent) had S. pneumoniae infection. In addition to cultures of blood and sputum the most useful initial tests were for sputum pneumococcal antigen and serum mycoplasma specific IgM. Twenty-six patients (5.7 per cent) died, seven within 48 h of admission. Multivariate analyses showed age, absence of chest pain, absence of vomiting, previous treatment with digoxin, tachypnoea, diastolic hypotension, confusion, leucopenia, leucocytosis, and raised blood urea levels were significantly correlated with death. Patients had a 21-fold increased risk of death if they had two of the following: admission respiratory rate greater than or equal to 30/min, admission diastolic BP less than or equal to 60 mmHg, urea greater than 7 mmol/l during admission. Mortality was not related to aetiology except that all three patients with combined Influenza A virus and Staphylococcus aureus infection died. Nine patients had legionella pneumonia; none died. No patients who died from pneumococcal pneumonia, mycoplasma pneumonia or staphylococcal pneumonia had received an appropriate antibiotic before admission. Such deaths are possibly preventable. Assisted ventilation was used in 22 patients of whom 14 survived. Hospital stay in survivors averaged 10.8 days; after six weeks 79 per cent were fit for normal activities, and 55 per cent showed resolution of radiographic signs of pneumonia. We recommend that antibiotics should be given as early as possible and chosen always to cover S. pneumoniae, and in addition M. pneumoniae during outbreaks, and S. aureus during influenza epidemics.