Pneumocystis jirovecii is a common cause of life-threatening pneumonia among immunocompromised patients. Using 400 fresh bronchoalveolar lavage samples, we compared prospectively routine direct immunofluorescence assay (DFA) and a real-time PCR assay, performed on a LightCycler system, for the detection of P. jirovecii. Among the 66 PCR positive samples, 31 were positive by DFA. No patient was found as having the pattern "PCR--ve/DFA+ve". The semi-quantification of the P. jirovecii DNA was represented by the cycle threshold (Ct). Using DFA as the gold standard, the sensitivity of the PCR was 100% for Ct > or = 28 and the specificity was 100% for Ct < 22. Between these two points, the results could be discrepant. The patients of the "22 < or = Ct < 28" group presented more frequently with a radiological interstitial syndrome than the "Ct > or = 28" group, and presented less frequently with HIV-infection and elevated lactate dehydrogenase (LDH) assay than in the "Ct < 22" group. A negative PCR allowed us to exclude the P. jirovecii pneumonia. The real-time PCR assay seems to be an accurate diagnosis method and could replace the DFA. The semi-quantitative results should be helpful to distinguish colonized, subclinically infected and P. jirovecii pneumonia patients.