ABSTRACT Background Syphilis continues to be a public health problem and its diagnosis still has limitations. Molecular diagnosis provides an alternative for rapid and effective management. The objective is to determine the accuracy of tests in the molecular diagnosis of syphilis. Methods We searched PubMed and Web of Sciences for articles related to molecular detection of syphilis during January 01, 2009, to December 31, 2019. The bivariate Reitsma model and the hierarchical Receiver Operating Characteristic (ROC) curve model were used to evaluate the diagnostic performance of molecular tests at a 95% confidence interval (CI). A subgroup meta-analysis was performed to explore sources of heterogeneity. Results 47 articles were identified for qualitative synthesis, of which 23 met the inclusion criteria for meta-analysis. The pooled sensitivities in conventional PCR (Polymerase chain reaction) and real-time PCR were 77.52 (59.50-89.01) and 68.43 (54.96-79.39) respectively. The pooled specificities were 98.00(90.73-99.59) and 98.84(97.55-99.46) respectively. Ulcer samples had a better performance (sensitivity of 79.88 (69.00- 87.62)) and specificity of 98.58 (97.25-99.27) and the major target genes were Polymerase A gene and tpp47 gene. Conclusion Our work showed that conventional PCR was more widely used than real-time PCR in the diagnosis of syphilis and ulcers were the best specimens. Sample types and target genes are factors that may influence the quality of the different tests. These results could provide evidence for further work in the direction of providing a more efficient diagnostic test.