We study selection into a voluntary unemployment insurance (UI) scheme. To disentangle behavioral effects from selection, we exploit variation in the sign-up induced by an early retirement scheme embedded in the UI system. Using Danish register data, we quantify the selection with an event study approach. Workers who sign up for UI are negatively selected in terms of subsequent unemployment. Different types of selection effects appear to be relevant: heterogeneity, risk-based selection and selection on moral hazard. The heterogeneity amounts to about one percentage point, while the predicted increase in unemployment is about three quarters of a percentage point from a baseline of about 5 percent. (Joint with Stefan Hochguertel.)