Objective: To examine the association between neonatal birth weight and maternal type 2 diabetes (T2DM), so as to provide insights into the formulation of the early T2DM prevention and improvements of maternal and children health. Methods: Single-nucleotide polymorphisms (SNPs) were collected from two genome-wide association studies (GWAS) that examined the association between neonatal birth weight and maternal T2DM. Inverse variance weighted method was employed for forward Mendelian randomization with 26 birth weight-associated SNPs as instrumental variables and maternal T2DM as the study outcome and for reverse Mendelian randomization with 18 maternal T2DM-associated SNPs as instrumental variables and maternal effects of neonatal birth weight as the study outcome. The heterogeneity was examined using Cochran's Q test, and the pleiotropy was evaluated using MR-Egger regression, while the robustness of the results was evaluated using leave-one-out test. Results: Cochran's Q test detected heterogeneity across two studies (P=0.019, 0.038), and random effect models were employed. Mendelian randomization showed that an increase in neonatal birth weight by per standard error (approximately 488 g) resulted a 29.9% reduction in the risk of maternal T2DM (OR=0.701, 95%CI: 0.547-0.899), and maternal T2DM increased the neonatal birth weight by 0.074 standard errors (OR=1.074, 95%CI: 1.043-1.106). No horizontal pleiotropy was seen for instrumental variables (P=0.241, 0.188). With each SNP excluded in turn, the results of Mendelian randomization study were robust. Conclusion There are bidirectional associations between neonatal birth weight and risk of maternal T2DM.