Objective: To determine the accuracy with which morphology alone can distinguish adenocarcinoma and squamous cell carcinoma in non-small cell lung cancer. Methods: We performed a systematic review and meta-analysis. Three data bases (MEDLINE, EMBASE, Scopus) were searched for studies on the diagnostic accuracy of subtyping non-small cell lung cancer. Accuracy data was abstracted and synthesized using bivariate mixed effects logistic regression as implemented in the midas package in Stata 14. Heterogeneity was assessed using the Higgins I2. Results: We included 17 studies (2,235 cases). Most studies had a low risk of bias. The pooled diagnostic accuracy for cytological diagnosis of adenocarcinoma resulted in a sensitivity of 63% (48–76%) and specificity of 95% (87–98%). The I2 values were 93 and 88% for sensitivity and specificity, respectively. The pooled diagnostic accuracy for the cytological diagnosis of squamous cell carcinoma resulted in a sensitivity of 84% (79–88%) and a specificity of 90% (84–94%). The I2 values were 69 and 86% for sensitivity and specificity, respectively. Conclusion: Accuracy varies widely by study and summary estimates do not provide a useful representation of accuracy. Squamous cell carcinoma was diagnosed more accurately than adenocarcinoma.