This paper examines the application of Information Theory to hospital discharge data. Information Theory offers a general methodology to compare sets of casemix proportions as a measure of (1) the concentration of admissions across hospitals for specific medical conditions and (2) specialization across diagnostic categories for individual hospitals. Unfortunately, Information Theory indices are difficult to interpret and subject to a potentially serious statistical bias when computed from discrete frequency counts, such as those obtained from discharge abstract data. The analysis presented here first clarifies the interpretation of Information Theory indices by relating them to formal statistical tests of hypotheses about hospital and diagnosis-specific patterns of admissions. It then documents the magnitude of the bias due to calculating indices from discrete frequency counts and proposes analytical strategies for dealing with this bias. Finally, the paper examines the empirical importance of the bias and the proposed adjustment, using data that are typical of those available for research on hospital casemix.