This paper considers the long-term effects of smoking on disability retirement in Sweden. Smoking is known to have damaging effects on health, but there is limited evidence on how the effects of smoking translate into worse labour market outcomes, such as the inability to work. In contrast to the few previous studies on smoking and disability retirement, we use a large population sample with registry information on smoking, which is recorded for all women who give birth in Sweden. Thanks to these comprehensive data, we are able to account for a much broader range of potential confounders. In particular, by the use of sibling and twin fixed effects, we account for unobserved heterogeneity in childhood environment and family characteristics. Given that smoking is often initiated in adolescence, one would suspect such factors to play important roles. Among individuals aged 50-64 in 2011, a simple model suggested smokers to have a 5 percentage point higher probability of receiving (full) disability pension, making them more than twice as likely as non-smokers to receive this. However, in a model with sibling fixed effects, the size of the effect was reduced by more than a third. The results point to the importance of confounders, such as childhood circumstances or behaviours, which were not accounted for by previous studies. We also consider effects on disability due to different health conditions. In relative terms, effects are the largest for circulatory conditions and tumours. Results are largely driven by health problems severe enough to merit hospitalization, and there is no evidence of a role played by financial incentives.