Amplification of Bartonella henselae DNA has been proposed as a diagnostic test for cat scratch disease (CSD). The sensitivities of the following three PCR assays were compared. PCR/rRNA with universal primers amplifies part of the 16S rRNA gene, followed by hybridization with a specific B. henselae probe; PCR/CS and PCR/HSP amplify portions of the gltA and the htrA genes, respectively, each followed by restriction fragment length polymorphism analysis. The threshold of detection of B. henselae DNA in pus was 10-4, 10-3, and 10-2 ng for PCR/rRNA, PCR/CS, and PCR/HSP, respectively. By these three assays, B. henselae DNA was detected in 100, 94, and 69% of 32 pus and lymph node specimens from CSD patients, respectively. The similar sensitivities of the PCR/rRNA and the PCR/CS assays for detecting B. henselae DNA in clinical specimens are in contrast to the 10-fold difference in sensitivities in favor of PCR/rRNA demonstrated with purified B. henselae DNA in sterile pus, suggesting that in the majority of cases, the bacterial load in clinical specimens is large enough to be identified by the PCR/CS assay. A two-step approach is suggested to achieve maximal sensitivity for detecting B. henselae in clinical specimens: initial testing by PCR/CS (which does not require hybridization), followed by PCR/rRNA with PCR/CS- negative specimens when CSD is strongly suspected.