Abstract Background This meta-analysis examines the comparative diagnostic performance of polymerase chain reaction (PCR) for the diagnosis of Pneumocystis pneumonia (PCP) on different respiratory tract samples, in both human immunodeficiency virus (HIV) and non-HIV populations. Methods A total of 55 articles met inclusion criteria, including 11 434 PCR assays on respiratory specimens from 7835 patients at risk of PCP. QUADAS-2 tool indicated low risk of bias across all studies. Using a bivariate and random-effects meta-regression analysis, the diagnostic performance of PCR against the European Organisation for Research and Treatment of Cancer–Mycoses Study Group definition of proven PCP was examined. Results Quantitative PCR (qPCR) on bronchoalveolar lavage fluid provided the highest pooled sensitivity of 98.7% (95% confidence interval [CI], 96.8%–99.5%), adequate specificity of 89.3% (95% CI, 84.4%–92.7%), negative likelihood ratio (LR−) of 0.014, and positive likelihood ratio (LR+) of 9.19. qPCR on induced sputum provided similarly high sensitivity of 99.0% (95% CI, 94.4%–99.3%) but a reduced specificity of 81.5% (95% CI, 72.1%–88.3%), LR− of 0.024, and LR+ of 5.30. qPCR on upper respiratory tract samples provided lower sensitivity of 89.2% (95% CI, 71.0%–96.5%), high specificity of 90.5% (95% CI, 80.9%–95.5%), LR− of 0.120, and LR+ of 9.34. There was no significant difference in sensitivity and specificity of PCR according to HIV status of patients. Conclusions On deeper respiratory tract specimens, PCR negativity can be used to confidently exclude PCP, but PCR positivity will likely require clinical interpretation to distinguish between colonization and active infection, partially dependent on the strength of the PCR signal (indicative of fungal burden), the specimen type, and patient population tested.