The number of human deaths due to rabies is currently underestimated to be 55,000 deaths per year. Biological diagnostic methods for confirmation of rabies remain limited, because testing on postmortem cerebral samples is the reference method, and in many countries, sampling brain tissue is rarely practiced. There is a need for a reliable method based on a simple collection of nonneural specimens. A new reverse-transcription, heminested polymerase chain reaction (RT-hnPCR) protocol was standardized at 3 participating centers in Cambodia, Madagascar, and France. Fifty-one patients from Cambodia, Madagascar, Senegal, and France were prospectively enrolled in the study; 43 (84%) were ultimately confirmed as having rabies. A total of 425 samples were collected from these patients during hospitalization. We studied the accuracy of the diagnosis by comparing the results obtained with use of biological fluid specimens (saliva and urine) and skin biopsy specimens with the results obtained with use of the standard rabies diagnostic procedure performed with a postmortem brain biopsy specimen. The data obtained indicate a high specificity (100%) of RT-hnPCR and a higher sensitivity (>/=98%) when the RT-hnPCR was performed with skin biopsy specimens than when the test was performed with fluid specimens, irrespective of the time of collection (i.e., 1 day after the onset of symptoms or just after death). Also, a sensitivity of 100% was obtained with the saliva sample when we analyzed at least 3 successive samples per patient. Skin biopsy specimens should be systematically collected in cases of encephalitis of unknown origin. These samples should be tested by RT-hnPCR immediately to confirm rabies; if the technique is not readily available locally, the samples should be tested retrospectively for epidemiological purposes.