The quality of medical records data for patients who were hospitalized with community-acquired pneumonia was assessed by comparing medical records data with data obtained in a prospective study of pneumonia for the period April 1, 1984, to December 31, 1984. One hundred five patients fulfilled the case definition of pneumonia for entry into the prospective study. One hundred twenty-seven patients were identified by medical records data. Seventy-three of the patients appeared in both studies. The positive predictive accuracy of the medical records data was 57%. When the etiologic diagnoses for the 73 patients identified by both studies were compared, there was agreement only 52.6% of the time. Streptococcus pneumoniae was overdiagnosed, and Mycoplasma, specific viral causes, and Haemophilus influenzae were not recorded by the medical records data. The quality of medical records data regarding pneumonia can be improved by changing the current ICD-9-CM coding system for pneumonia and by providing instruction and an algorithm for abstractors to follow in assigning a diagnosis of pneumonia.