The association between polymorphism of DNA methyltransferases 3B and cancer risk has been widely studied recently, and no consensus conclusion is available up to now. We perform a comprehensive search using the databases of Medline, ISI Web of Knowledge and Embase. The odds ratio (OR) and its 95% confidence interval (95% CI) are used to investigate the strength of the association. A total of 24 case–control studies with 15,647 individuals are included in this meta-analysis. For −149C > T (17 studies, 5229 cases and 6910 controls), no evidence indicate that individuals carrying the variant genotypes (CC + CT), relative to those carrying the wild homozygote TT genotype, have an increased risk of cancer (OR = 1.03; 95% CI = 0.84–1.26; P = 0.76). Similarly, no cancer risk is found in the subgroup analyses. For −579G > T (11 studies, 3513 cases and 3714 controls), significantly decreased risks of cancer are observed, and the ORs (95% CI) are 0.70 (0.56–0.87) for GT versus TT, 0.70 (0.57–0.85) for GG + GT versus TT and 0.76 (0.63–0.93) for G-allele versus T-allele, respectively. Subgroup analyses stratified by ethnicity and types of cancer are also performed, and results indicated that −579G > C polymorphism is associated with risk of cancer in Asians [0.68 (0.53–0.87) for GT vs. TT] but not in Europeans [0.82 (0.63–1.07) for GT vs. TT]. We also observe that the −579G is associated with decreased risk of colorectal cancer [0.49(0.38–0.62) for GT vs. TT]. More studies with larger sample size were needed to provide more precise evidence.