Serum albumin was measured in 126 men (average age 70.6; range 40 to 96) of a Veterans Administration Nursing Home, and was correlated with other items in an extensive clinical data base, including death or survival during the year after the analysis. The reason for institutionalization was chronic neurologic disease or other disabling physical condition in 63 men (group A), and psychiatric disorder in 63 men (group B). In group A, the proportions of men with albumin less than 3.5, 3.5-4.0, and greater than 4.0 g/dl were 6%, 37%, and 57%, respectively. In this group, the serum albumin level was significantly (p less than 0.05) correlated with death rate, hemoglobin, hematocrit, serum cholesterol, and serum lactic dehydrogenase. The death rate in group A during the year after the albumin analysis was 25%. For the patients with albumin level less than 3.5, 3.5-4.0, and greater than 4.0 g/dl, the death rates were 50%, 43%, and 11% respectively (p less than 0.01 for comparison of the former two groups with the latter). The subgroup with albumin 3.5-4.0 g/dl represented only 37% of the men in group A, but accounted for 63% of the group's deaths. In group B, serum albumin level was not significantly correlated with any other clinical variable. Death rate during the year after the albumin analysis was only 2% in group B, and did not correlate with the albumin level. These data indicate that, in nonpsychiatric Nursing Home men, the desirable level for the serum albumin concentration is higher than 3.5 g/dl.