Individual choice among health insurance policies may result in risk-based sorting across plans. Such adverse selection induces three types of losses: efficiency losses from individuals' being allocated to the wrong plans; risk-sharing losses, because premium variability is increased; and losses from insurers' distorting their policies to improve their mix of insureds. We discuss the potential for these losses and present empirical evidence on adverse selection in two groups of employees: Harvard University and the Group Insurance Commission of Massachusetts (serving state and local employees). In both groups, adverse selection is a signiacant concern. Harvards decision to contribute an equal amount to all insurance plans led to the disappearance of the most generous policy within three years. The Group Insurance Commission has contained adverse selection by subsidizing premiums proportionally and managing the most generous policy very tightly. A combination of prospective or retrospective risk adjustment, coupled with reinsurance for high-cost cases, seems promising as a way to provide appropriate incentives for enrollees and to reduce losses from adverse selection.