The relationship between depression and severe non-alcoholic fatty liver disease (NAFLD) has not been clearly defined. We conducted a longitudinal cohort study and a two-sample Mendelian randomization (MR) analysis to assess the association of depression with severe NAFLD risk. We used individual data from the UK Biobank study with 481,181 participants, and summary data from published genome-wide association studies. The association between depression and severe NAFLD was assessed using Cox proportional hazards regression analysis. Two-sample MR for depression with NAFLD was conducted, the principal analysis employed the inverse variance weighted (IVW) approach. In the observational study, after a median follow-up of 13.46 years, 4,563 participants had severe NAFLD. In multivariable-adjusted model, participants with depression had an increased risk of severe NAFLD (hazards ratio:1.21, 95% confidence interval (CI):1.09-1.34), as compared to those without depression. In subgroup analyses, the association between depression and severe NAFLD risk was generally observed across different subgroups. For the MR, result also showed that genetically predicted depression was causally associated with a higher risk of NAFLD (odds ratio:1.55, 95%CI:1.10-2.19) in IVW. Our study revealed a prospective association of depression with severe NAFLD, thus potentially necessitating clinical monitoring of individuals with depression for severe NAFLD.