The prognostic significance of delayed therapy in Legionnaires' disease is poorly defined. Thirty-nine consecutive serologically confirmed cases of Legionnaires' disease were reviewed to examine whether an association exists between delayed therapy and prognosis. Clinical and laboratory factors predictive of mortality were also sought. Thirty-one cases (79%) were classified as having severe pneumonia at diagnosis. Thirty-six patients (92%) had community-acquired infection, and three patients (8%) had nosocomial disease. Ten patients died, resulting in a crude mortality rate of 26%. At the first assessment, variables noted for pneumonia associated with death were low diastolic blood pressure (p < 0.02), low serum albumin concentration (p < 0.04), and increased number of days from onset of pneumonia to hospitalisation (prodrome) (p < 0.02). However, multiple logistic regression analysis revealed that the prodrome was the only variable noted at diagnosis that achieved significance (p = 0.024). Mortality also correlated with both delay in the initiation of erythromycin therapy following admission (p < 0.001) and the total delay in starting erythromycin therapy (p < 0.001). It is therefore recommended that erythromycin be included early in the empiric therapy of severe community-acquired pneumonia.