With cancer being a major cause of death worldwide, microRNAs (miRNAs) have been investigated as novel and non-invasive biomarkers for cancer diagnosis. Recently, microRNA-21 (miR-21) attracts much attention for its aberrant expression and has been widely studied in various cancers. However, the inconsistent results from studies make it hard to evaluate the diagnostic value of miR-21 in cancer diagnosis, which lead us to conduct this meta-analysis. We conducted a comprehensive literature search in the Medline, Embase, PubMed, CNKI, and Web of Science before July 1, 2014. STATA 12.0 software was used for calculation and statistical analysis. The pooled sensitivity, specificity, positive and negative likelihood ratio (PLR, NLR), and diagnostic odds ratio (DOR) were used to assess the diagnostic performance of miR-21 for cancers. Seventy-three studies in 60 articles were involved in this meta-analysis, with a total of 4684 patients with cancer and 3108 controls. The overall parameters were calculated from all the included studies: sensitivity of 0.78 (95% confidence interval (CI) 0.74-0.81), specificity of 0.83 (95% CI 0.80-0.86), PLR of 4.5 (95% CI 3.8-5.4), NLR of 0.27 (95% CI 0.23-0.32); DOR of 17 (95% CI 12-23), and area under the curve (AUC) of 0.88 (95% CI 0.84-0.90). In addition, we performed subgroup analyses based on ethnicity, cancer types, and sample types. Results from subgroup analysis showed that cancer types and sample types were the sources of heterogeneity in our meta-analysis. The overall diagnostic value of miR-21 is not very high for cancer diagnosis; however, it is affected significantly by the types of cancer and specimen. MiR-21 has a relatively high diagnostic value for detecting breast cancer, and miR-21 assays based on plasma, serum, and tissue achieved relatively higher accuracy.