We assessed the performance of a duplex real-time PCR assay for bla(KPC) and bla(NDM) performed directly (D-PCR) on perianal and perirectal swabs and stool. Spiked specimens and 126 clinical surveillance swabs (comprising a sensitivity panel of 46 perirectal double swabs previously determined to be culture positive for bla(KPC)-PCR-positive Enterobacteriaceae and a specificity panel of 80 perianal swabs from patients at risk of carbapenemase-producing Enterobacteriaceae [CPE] colonization) were studied. For the surveillance swabs, D-PCR was compared to PCR after broth enrichment (BE-PCR) and two culture-based methods: the HardyCHROM ESBL agar (HC-A) and the CDC screening (CDC-A) methods. PCR was performed on morphologically distinct colonies that were isolated by culture. All of the initial PCR testing was done without extraction using a simple lysis procedure. The analytical sensitivities of D-PCR for bla(KPC) were 9 CFU/μl (for swabs) and 90 CFU/μl (for stool), and for bla(NDM), it was 1.9 CFU/μl (for both swabs and stool). In the clinical sensitivity panel, D-PCR and BE-PCR were initially positive for bla(KPC) in 41/46 (89.1%) and 43/46 (93.5%) swabs, respectively. The swabs that were initially negative by D-PCR (n = 5) and BE-PCR (n = 3) were visibly stool soiled; all swabs were bla(KPC) positive upon repeat testing after lysate extraction. The CDC-A and HC-A yielded bla(KPC)-positive Enterobacteriaceae from 36/46 (78.3%) and 35/46 (76.1%) swabs, respectively (sensitivities of D-PCR/BE-PCR postextraction of soiled specimens versus HC-A, P = 0.0009, and versus CDC-A, P = 0.0016). All swabs in the specificity panel were negative for CPE by all four methods. D-PCR allows for the timely detection of bla(KPC) and bla(NDM) carriage with excellent sensitivity when specimens visibly soiled with stool undergo preparatory extraction.