Tumor tissues are often absent or insufficient for testing epidermal growth factor receptor (EGFR) mutations to guide EGFR tyrosine kinase inhibitors (TKIs) treatment of patients with nonsmall cell lung cancer (NSCLC). We conducted this systematic review and meta-analysis to assess whether blood can be used as a substitute for tumor tissue in detecting EGFR mutations. MEDLINE, EMBASE, and the Cochrane Library were searched for studies that provided data to estimate the accuracy of blood testing against tissue testing in NSCLC patients and/or those directly compared the efficacy of EGFR TKIs in EGFR mutant and wild-type patients according to sources of specimens. Sensitivity, specificity, and concordance rate were used as measures of the accuracy. Risk ratio (RR) for objective response and hazard ratio (HR) for progression-free survival (PFS) and overall survival (OS) were used as measures for treatment efficacy. We combined the effects by using the fixed-effects model unless there was evidence of heterogeneity, in which case a random-effects mode was used. This systematic review included 25 studies with 2605 patients. The pooled overall sensitivity, specificity, and concordance rate were 0.61, 0.90, and 0.79, respectively. Serum showed lower sensitivity (0.56 vs 0.65) but higher specificity (0.95 vs 0.85) and higher concordance (0.86 vs 0.74) than plasma. EGFR mutations (exon 19 or 21) in blood were significantly associated with objective response (RR: 4.08; 95% confidence interval [CI] 2.48-6.70), PFS (HR: 0.72; 95% CI 0.64-0.80), and OS (HR: 0.71; 95% CI 0.50-0.99). Importantly, the association of the mutations with the 3 clinical outcomes for serum was similar to that for tumor tissue and higher than that for plasma. Blood, in particular serum, is a good substitute when tumor tissue is absent or insufficient for testing EGFR mutations to guide EGFR TKIs treatment in patients with NSCLC. EGFR mutation positivity in blood could be used to recommend EGFR TKIs treatment, but the absence of blood positivity should not necessarily be construed with confirmed negativity.