During the 1990s, a new subsidized early retirement option was assigned quasi-randomly to two-thirds of Norwegian elderly workers. We use this 'natural experiment' to evaluate how economic incentives affect retirement behaviour. The new retirement option reduced employment substantially, and this effect escalated over time. It did not substitute for disability pension or long-term unemployment, and approximately two out of three retirees would have stayed employed without it. Subsidized retirement was primarily voluntary, but there is also some evidence that firms took advantage of the retirement programme to 'push out' excess workers.