Obesity has been implicated in the aetiology of liver disease. However, to date, evidence is largely drawn from cross-sectional studies, where interpretation is hampered by reverse causality, and from studies on clinical populations that have limited generalisability. In this prospective cohort study, data on body mass index (BMI) and covariates were collected at baseline on 18 863 male government employees (aged 40-69 years). Respondents were then followed up for a maximum of 38 years of age. Mortality surveillance gave rise to 13 129 deaths, 122 of which were due to liver disease (57 cancers; 65 non-cancers). In age-adjusted analyses, BMI was positively related to total liver disease mortality (hazards ratio per 1 s.d. increase in BMI; 95% confidence interval (CI): 1.36; 1.14, 1.62) in a graded fashion across the weight categories (P-value for trend: 0.01). The magnitude of this association was somewhat stronger for non-cancer liver disease deaths (1.47; 1.16, 1.86) than for cancer liver disease deaths (1.25; 0.96, 1.62). Excluding deaths in the first 10 years of follow-up somewhat strengthened the BMI-non-cancer liver disease association. Adjustment for socioeconomic position, other candidate confounders and mediating factors led to the modest attenuation of these associations. Further investigation in prospective cohort studies with more detailed data on liver disease, for instance using biochemical tests of liver function or hepatic ultrasonography, is warranted.