The issue of selection bias was investigated using data from 22 HMOs who are enrolling Medicare beneficiaries under Tax Equity and Fiscal Responsibility Act of 1982 (TEFRA) at-risk contracts. The study differs from previously published analyses of this issue in that it deals with the current Medicare risk program (TEFRA) rather than with earlier Demonstration Programs; as an indicator of selection bias, it utilizes beneficiary functional health status at enrollment; and it examines selection not only at the mean of the health status distribution, but at the two tails (very disabled, very able) as well. For each of the participating HMOs, the functional health status of recent Medicare enrollees was compared with that of a control group of randomly chosen fee-for-service beneficiaries. None of the HMOs experienced adverse selection, whether measured in terms of overall (mean) health status of enrollees or in terms of the proportion of the very disabled population that chose to join. Nine of the 22 HMOs were considered to have experienced favorable selection on the basis of the mean health status of new enrollees. In addition, ten more HMOs were found to have experienced favorable selection in one or both tails of the health status distribution. Although a specific cause for the observed enrollment patterns is not identified, speculation is made on factors that may or may not contribute. Evidence suggests that beneficiary self-selection is probably a more important explanation of these patterns than purposeful actions of HMOs to discourage enrollment by sicker beneficiaries (i.e., "skimming").