OBJECTIVE. Depression is a common co-morbidity of type 2 diabetes. However, the causality and underlying mechanisms remain unclear. RESEARCH DESIGN AND METHODS. We applied bi-directional Mendelian randomization (MR) to assess causality between type 2 diabetes and self-reported depression. Using the UK biobank, we performed 1) GWAS, separately, and 2) multi-phenotype GWAS (MP–GWAS) of type 2 diabetes (cases=19,344, controls=463,641) and depression, using two depression definitions–clinically diagnosed major depressive disorder (MDD, cases=5,262, controls=86,275) and self–reported depressive symptoms (PHQ-9, n=153,079). The FinnGen study was used for replication for MDD (n=23,424) and type 2 diabetes (n=32,469). Based on the results, we analyzed expression quantitative trait loci (eQTL) data from public databases to identify target genes in relevant tissues. RESULTS. MR demonstrated a significant causal effect of depression on type 2 diabetes (OR=1.18[1.06–1.32], p=0.0024), but not in the reverse direction. GWAS of type 2 diabetes and depressive symptoms did not identify any shared loci between them, whereas MP-GWAS identified seven shared loci mapped toTCF7L2, CDKAL1, IGF2BP2, SPRY2, CCND2-AS1, IRS1, CDKN2B-AS1. MDD did not yield genome-wide significant loci in either GWAS or MP-GWAS. We found that most MP-GWAS loci had an eQTL, including SNPs implicating the cell cycle gene CCND2 in pancreatic islets and brain, and key insulin signaling gene IRS1 in adipose tissue, suggesting a multi-tissue and pleiotropic underlying mechanism. CONCLUSION. Our study reveals the complexity in the depression-diabetes relationship and our results have important implications for a more efficient prevention of type 2 diabetes from early adulthood when depressive symptoms usually occur.