ABSTRACT Aims To assess the importance of health selection in the association between unemployment and alcohol‐related mortality by comparing mortalities of those unemployed from workplaces experiencing different levels of downsizing. The more severe the downsizing, the less dependent unemployment would be on personal characteristics and the weaker the role of health selection. Design We estimated hazards models of unemployment on alcohol‐related diseases and poisonings and external causes with alcohol as a contributing cause over follow‐ups of 0–5, 6–10 and 11–20 years and at different levels of downsizing (stable, downsized or closed). Setting Finland, 1990 to 2009. Participants A register‐based random sample of employees aged 25–63 in privately owned workplaces (n = 275 738). Measurements The outcome was alcohol‐related death and the exposure was unemployment. We adjusted for age, sex, year, education, marital status, health status, workplace tenure, industry, region and unemployment rate. Findings Alcohol‐disease mortality was elevated among the unemployed throughout the follow‐up, regardless of the level of downsizing. At 11–20 years after baseline, those unemployed from stable workplaces had a 2.46 hazard ratio (HR) (95% confidence interval CI = 2.14–2.82), those from downsized workplaces 1.94 (95% CI = 1.64–2.30) and those from closed workplaces 2.13 (95% CI = 1.75–2.59), when compared with the controls. Alcohol‐related external‐cause mortality at 0–5 years follow‐up was only associated with unemployment from stable workplaces (HR = 1.39, 95% CI = 1.22–1.58), but over time, an association emerged among those unemployed following downsizing and closure. At 11–20‐year follow‐up, the HR following downsizing was 1.83 (CI 95% = 1.37–2.45) and 1.54 (95% CI = 1.03–2.28) following closure. Conclusions There is some indication that alcohol‐related ill‐health may lead to unemployment in Finland. However, the persistent long‐term association between unemployment and alcohol‐related mortality even after workplace closure may imply a causal relation.