The interest in measuring health status with survey instruments has not been matched with an analysis of their performance characteristics in the field. We used the Medical Outcome Study Short Form (MOS-20) to assess health outcomes among patients who were hospitalized in one of two public hospitals. We mailed the MOS-20 and a series of transition questions, which asked about changes in health, to patients admitted in the previous year. We received 414 completed surveys from 480 patients at baseline and follow-up data on 90% of these patients six months later. Baseline MOS-20 scores for study patients were significantly lower, corresponding to worse health, than previously reported outpatient and general population cohorts. While the direction of change on serial applications of the MOS-20 paralleled the patients' perception of change reported on transition questions, many patients who reported their health had become worse also recorded the lowest possible score on the MOS-20 at baseline. These low baseline MOS-20 scores prohibited the recognition of larger declines in function during the follow-up period. This floor in the response range creates an instrument bias against documenting a decline in health among severely ill patients, the group in which it may be most important to detect such a change.