This study aims to learn whether the annual clinical and laboratory screening of nursing home residents provides significant information about their chance of dying during the following year. In August 1984, a comprehensive clinical data base was compiled for 176 male residents of this VA nursing home. During the next 14 months, 12 men were discharged to other locations and were dropped from the study; among the remaining 166, who comprised the study group of this report, 24 died. The most common immediate causes of death were infections (67%) and cardiac disorders (25%). Twenty of the deaths occurred after transfer to the acute hospital services. Among the 67 items in the clinical data base (including absence or presence of 17 diagnoses and 16 drugs), eight were significantly correlated with death rate. Age and functional impairment were directly related, and inversely related were the following: body weight as percent of ideal, triceps skin fold, hematocrit, hemoglobin, serum albumin, and serum cholesterol. Multivariate analysis showed cholesterol and hematocrit to be the most informative of the eight mortality predictors and to correlate with death independently of age and functional level. Subgroups defined on the basis of combinations of mortality-related attributes differed many fold in their death rates. For example, men with cholesterol less than or equal to 156 mg/dl and hematocrit less than or equal to 41% died at a rate 42 times the rate of men with values above both thresholds. For each mortality-related attribute, death rate varied with the level of the attribute. This relationship extended into the generally accepted "normal ranges" for cholesterol, hematocrit, hemoglobin, and albumin.