To determine the prevalence and incidence of and the relative risks for positive tuberculin skin tests among employees of a large, urban teaching hospital. Retrospective cohort study. Barnes Hospital, St. Louis, Missouri. Hospital personnel employed at any time between January 1989 and July 1991. 684 of 6070 employees screened (11.3% [95% CI, 10.4% to 12.1%]) had positive tuberculin skin tests. Factors associated with a positive result were age (odds ratio, 2.02 per decade [CI, 1.87 to 2.18]; P < 0.0001); black race (odds ratio, 1.58 [CI, 1.26 to 2.00]; P < 0.0001); Asian race (odds ratio, 16.7 [CI, 9.33 to 29.9]; P P P = 0.001). Twenty-nine of 3106 employees who had at least two tests had skin-test conversions (0.93% [CI, 0.60% to 1.3%]); 15 of these conversions (52%) occurred among employees who had no direct contact with patients. Only the percentage of low-income persons within the employee's residential postal zone (odds ratio 1.39 [CI, 1.09 to 1.78]; P = 0.0075) was independently associated with conversion. The most important associations with a positive tuberculin skin test were older age, minority group status, and the proportion of low-income persons within the employee's residential postal zone. Skin-test conversion was independently associated only with the percentage of low-income persons in the employee's postal zone. Stratifying employees according to degree of contact with patients or according to departmental group was not useful in determining risk for a positive tuberculin skin test or for skin-test conversion. For certain groups of employees, an exposure to tuberculosis in the community probably poses a greater risk than exposure in the hospital setting.