An expanded respiratory isolation policy was implemented in a public hospital that cares for about 200 patients with active tuberculous each year. This led to proper isolation of > or = 95% of patients with tuberculosis on admission but involved an 8-fold overuse of isolation rooms. We developed a model policy to decrease overisolation of nontuberculous patients. Clinical findings in 295 patients admitted to respiratory isolation during a 3-mo period were evaluated for their usefulness in determining which patients had tuberculosis. Multivariate analysis identified five predictive variables: chest radiograph with upper lobe infiltrate (odds ratio, 5.00; CI, 2.38 to 10.51; p = 0.001) or cavity (odds ratio, 3.93; CI, 1.06 to 14.62; p = 0.041), history of having known someone with tuberculosis (odds ratio, 2.42; CI, 1.10 to 5.32, p = 0.027), self-reported positive tuberculin skin test (odds ratio, 5.67; CI, 1.57 to 22.01; p = 0.009), self-reported isoniazid preventive therapy (odds ratio, 0.18; CI, 0.04 to 0.82; p = 0.027). Using these variables to determine which patients required isolation would have decreased the number of isolated nontuberculous patients from 253 to 95, but it would have missed eight of 42 patients with tuberculosis. Further work is needed to identify clinical predictors that would decrease overuse of isolation beds while maintaining satisfactory sensitivity for patients with tuberculosis.