Background The associations between Rad51 gene polymorphisms (G135C and G172T) and risk of cancer have been investigated, but the results were inconclusive. To get a comprehensive evaluation of the association above, we performed a meta-analysis of published studies. Methods A computerized search of PubMed, Embase and Web of Knowledge databases for all relevant studies was performed and the data were analyzed in a meta-analysis. The overall odds ratio (OR) with the 95% confidence interval (95% CI) was calculated to assess the strength of the association between Rad51 polymorphisms and cancer risk. Data were analyzed using fixed- or random-effects model when appropriate. Sensitivity analysis and publication bias test were also estimated. Results Overall, a total of 54 case-control studies were included in the current meta-analysis, among which 42 studies with 19,142 cases and 20,363 controls for RAD51 G135C polymorphism and 12 studies with 6,646 cases and 6,783 controls for G172T polymorphism. For G135C polymorphism, the pooled results indicated that significantly increased risk was found in overall cancers (homozygote model: OR = 1.776, 95% CI = 1.288–2.449; allelic genetic model: OR = 1.169, 95% CI = 1.016–1.345; recessive model: OR = 1.946, 95% CI = 1.336–2.835), especially in breast cancer (homozygote model: OR = 1.498, 95% CI = 1.026–2.189; recessive model: OR = 1.732, 95% CI = 1.170–2.562). For G172T polymorphism, a decreased cancer risk was observed in head and neck cancer (homozygote model: OR = 0.621, 95% CI = 0.460–0.837; allelic genetic model: OR = 0.824, 95% CI = 0.716–0.948; recessive model: OR = 0.639, 95% CI = 0.488–0.837). Conclusions Our results suggested that the Rad51 G135C polymorphism is a candidate for susceptibility to overall cancers, especially to breast cancer, and that the Rad51 G172T might play a protective role in the development of head and neck cancer.