Diabetes mellitus (DM) is widely considered to be associated with the risk of diverse cancers; however, the association between DM and the risk of leukemia is still controversial. Thus, a detailed meta-analysis of cohort studies was conducted to elucidate this association. Eligible studies were screened through the electronic searches in PubMed, Web of Science, and Embase from their inception to August 11, 2020. Summary relative risks (RRs) and 95% confidence intervals (CIs) were computed through the random-effects model. Eighteen articles involving 10,516 leukemia cases among a total of 4,094,235 diabetic patients were included in this meta-analysis. Overall, twenty-five RRs were synthesized for type 2 diabetes mellitus (T2DM) and yielded a summary RR of 1.33 (95%CI, 1.21–1.47; p < 0.001). For type 1 diabetes mellitus (T1DM), 7 RRs were combined, however, the pooled RR was insignificant (RR, 1.08; 95%CI, 0.87–1.34; p = 0.48). Interestingly, the summary RR for East Asia (RR, 1.83, 95%CI, 1.63–2.06) was significantly higher than that for Europe (RR, 1.11, 95%CI, 1.06–1.15), Western Asia (RR, 1.40, 95%CI, 1.25–1.54), North America (RR, 1.14, 95%CI, 1.08–1.20), and Australia (RR, 1.47, 95%CI, 1.25–1.71). Moreover, we found that patients with a shorter T2DM duration (1–5 years) had a higher risk of leukemia compared to those with a longer duration (5.1–10 years). Overall, this meta-analysis suggests there is a moderately increased risk of leukemia among T2DM patients, but not in T1DM patients. Further investigation is warranted.