Invasive candidiasis (IC) is a significant cause of morbidity and mortality. Diagnosis relies on culture-based methods, which lack sensitivity and delay diagnosis. We conducted a systematic review assessing the diagnostic accuracy of PCR-based methods to detect Candida spp. directly in blood samples. We searched electronic databases for prospective or retrospective cohort and case-control studies. Two reviewers abstracted data independently. Meta-analysis was performed using a hierarchical logistic regression model. Random-effects metaregression was performed to assess the effects of study methods and infection characteristics on sensitivity or specificity values. We included 54 studies with 4,694 patients, 963 of whom had proven/probable or possible IC. Perfect (100%) sensitivity and specificity for PCR in whole-blood samples was observed when patients with cases had candidemia and controls were healthy people. When PCR was performed to evaluate patients with suspected invasive candidiasis, the pooled sensitivity for the diagnosis of candidemia was 0.95 (confidence interval, 0.88 to 0.98) and the pooled specificity was 0.92 (0.88 to 0.95). A specificity of >90% was maintained in several analyses considering different control groups. The use of whole-blood samples, rRNA, or P450 gene targets and a PCR detection limit of ≤ 10 CFU/ml were associated with improved test performance. PCR positivity rates among patients with proven or probable IC were 85% (78 to 91%), while blood cultures were positive for 38% (29 to 46%). We conclude that direct PCR using blood samples had good sensitivity and specificity for the diagnosis of IC and offers an attractive method for early diagnosis of specific Candida spp. Its effects on clinical outcomes should be investigated.