We aimed to assess the diagnostic value of the nuclear matrix protein 22 (NMP22) test in detecting bladder cancer. We searched public databases including PubMed, Springer, MEDLINE, Elsevier Science Direct, Google scholar and Cochrane Library published before September 2015. Cystoscopy and voided urine cytology (VUC) were golden standards. Sensitivity, specificity, positive likelihood ratio (LR), negative LR and diagnostic odds ratio (DOR) of NMP22 test from included studies were meta-analyzed. The summary receiver operating characteristic (SROC) curve was constructed, and the area under the curve (AUC) and an index Q* were summarized. Subgroup analyses were performed. In addition, Egger's test was used to detect publication bias. A total of 24 studies consisting of 8848 patients with bladder cancer were included in the present meta-analysis. The results of sensitivity, specificity, positive LR, negative LR and DOR of NMP22 test were 0.71 (95% CI = 0.69 to 0.72), 0.80 (95% CI = 0.79 to 0.81), 2.99 (95% CI = 2.42 to 3.71), 0.42 (95% CI = 0.35 to 0.50) and 7.45 (95% CI = 5.32 to 10.43) respectively. The AUC and Q* index were 0.7846 and 0.7225, respectively. Subgroup analysis suggested that cutoff value might be one source of heterogeneity. Egger's test showed that no publication bias existed (P > 0.05). The NMP22 test may be appropriate for detecting bladder cancer, but it cannot replace the cystoscopy and VUC in the clinical diagnosis. Further studies are needed to unify the cut-off value and evaluate the diagnostic efficiency of NMP22 test.