The authors discuss the objectives and definition of the Severity of Illness Index, which has been developed and refined at The Johns Hopkins University over the past 5 years. In addition, the training program for raters, the method used to ascertain reliability, and data from reliability testing in 18 hospitals are described. After at least 2 months' experience with severity scoring, the average agreement between hospital raters and the staff reliability rater varied between 90.8% and 97.7%, with an overall weighted average agreement of 93.5%. Several methods to evaluate the validity of the Index are presented. The conclusion is that the Severity of Illness Index is a reliable and valid tool for measuring inpatient severity of illness.