In most countries, the unemployed are entitled to unemployment benefits only if they have previously worked a minimum period of time. This institutional feature creates a sharp change at eligibility in the disutility from unemployment and may distort the duration of jobs. In this paper, we evaluate this eligibility effect using a regression discontinuity approach. Our evidence is based on longitudinal social security data from Portugal, where the unemployed are required to work a relatively long period to collect benefits. We find that monthly transitions from employment to unemployment increase by 10% as soon as the eligibility condition is met. This result is driven entirely by transitions to subsidised unemployment, which increase by 20%, as non-subsidised unemployment is not affected. The effects are even larger for the unemployed with high replacement ratios or those who meet the eligibility condition from multiple short employment spells.