This study examines whether the MedisGroups admission severity groups give unbiased estimates of 30-day mortality in 3037 Medicare-aged patients who were hospitalized in 1985 through 1986 with acute myocardial infarction. The average observed death rate for all acute myocardial infarction patients in the study who were in a given admission severity group was used to estimate the expected death probability for each case in a given group. (This is the same method used by the Pennsylvania Health Care Cost Containment Council for risk adjusting hospital mortality by diagnosis related groups in that state.) When compared with observed deaths, estimates of expected mortality were significantly biased for many patient attributes (eg, age, location of acute myocardial infarction, history of congestive heart failure, serum potassium level, serum urea nitrogen level, pulse rate, and blood pressure). These results are consistent with a conclusion that the MedisGroups scoring algorithm omits some important risk variables, inappropriately includes some other variables reflecting postadmission status, and gives the wrong weights to some appropriate risk variables. To the extent that these findings are also applicable to current MedisGroups scoring algorithms and to other conditions and procedures, MedisGroups admission severity groups cannot fairly adjust for interhospital case mix differences in outcome studies.