ObjectiveTo investigate the genetic association between nonalcoholic fatty liver disease （NAFLD） and type 2 diabetes mellitus （T2DM） using bidirectional two-sample Mendelian randomization （MR）， as well as the causal relationship between NAFLD and T2DM across different body mass index （BMI） categories. MethodsThe data were derived from genome-wide association studies conducted in European populations， with a sample size of 32 941 cases for NAFLD， 312 646 cases for T2DM， and 681 275 cases for BMI. The univariate and multivariate MR methods were used to assess the bidirectional causal relationship between NAFLD and T2DM in the general population and across different BMI subtypes. The methods of inverse-variance weighting， MR-Egger regression， constrained maximum likelihood and model averaging， and weighted median were used to conduct the MR analysis， and MR-Pleiotropy Residual Sum and Outlier， radial MR， the MR-Egger intercept method， and the Cochrane Q test were used for sensitivity analysis. ResultsThe univariate MR analysis revealed a bidirectional causal relationship between NAFLD and T2DM in the general population （forward analysis： odds ratio ［OR］=9.75， 95% confidence interval ［CI］： 2.57‍ ‍—‍ ‍37.00， P<0.001； reverse analysis： OR=1.01， 95%CI： 1.00‍ ‍—‍ ‍1.01， P<0.01）. After adjustment for BMI， the multivariate MR analysis showed that the causal relationship between NAFLD and T2DM remained significant in the general population （OR=33.12， 95%CI： 7.57‍ ‍—‍ ‍144.95， P<0.000 1）. The subgroup analysis showed a causal relationship between NAFLD and T2DM across all BMI subtypes （lean subgroup： OR=12.19， 95%CI： 3.35‍ ‍—‍ ‍44.40， P<0.001； overweight subgroup： OR=4.30， 95%CI： 1.69‍ ‍—‍ ‍10.92， P<0.01； obese subgroup： OR=1.67， 95%CI： 1.14‍ ‍—‍ ‍2.44， P<0.01）. ConclusionThis study reveals the causal relationship between NAFLD and T2DM in the general population of NAFLD and across different BMI subtypes from a genetic perspective.