Background: Enrollment in Medicare health maintenance organization (HMOs) is encouraged because of the expectation that HMOs can help slow the growth of Medicare costs. However, Medicare HMOs, which are paid 95 percent of average yearly fee-for-service Medicare expenditures, are increasingly believed to benefit from the selective enrollment of healthier Medicare recipients. Furthermore, whether sicker patients are more likely to disenroll from Medicare HMOs, thus raising average fee-for-service costs, is not clear. Methods: We used Medicare enrollment and inpatient billing records for southern Florida from 1990 through 1993 to examine differences in the use of inpatient medical services by 375,406 beneficiaries in the Medicare fee-for- service system, 48,380 HMO enrollees before enrollment, and 23,870 HMO enrollees after disenrollment. We also determined whether these differences were related to demographic characteristics and whether the pattern of use after disenrollment persisted over time. Results: The rate of use of inpatient services in the HMO-enrollment group during the year before enrollment was 66 percent of the rate in the fee-for-service group, whereas the rate in the HMO-disenrollment group after disenrollment was 180 percent of that in the fee-for-service group. Beneficiaries who disenrolled from HMOs re-enrolled at about the time that their level of use dropped to that in the fee-for-service group. Conclusions: These data show marked selection biases with respect to HMO enrollment and disenrollment. These biases undermine the effectiveness of the Medicare managed-care system and highlight the need for longitudinal and population-based studies.