Standard microbiologic techniques were compared with a rapid diagnostic method based on the amplification by polymerase chain reaction (PCR) of a fragment of the IS6110 insertion element (present in multiple copies in the Mycobacterium tuberculosis genome) for the detection of M. tuberculosis in specimens obtained from children diagnosed as having primary tuberculosis on clinical grounds. Two (n = 7) or three (n = 15) gastric aspirates were obtained from the 22 children with primary tuberculosis. All specimens were negative for mycobacteria by acid-fast staining and culture. When DNA was purified from the clinical specimens and aliquots of each sample were amplified in duplicate, 15 of 59 (25%) specimens gave at least one positive result. Increasing beyond two the number of times that samples were tested did not appreciably improve sensitivity. Testing multiple samples from the same individual increased the diagnostic yield. Thus, when three different samples from the same subject were tested two times each, two or more positive results were obtained from 9 of 15 children with primary tuberculosis but 0 of 17 control subjects. Samples from children with symptoms, recent contact with patients with active tuberculosis, vesicular tuberculin responses, or abnormal chest radiographs were more frequently positive than those from patients whose only manifestation of tuberculosis was a positive (but not vesicular) tuberculin response. Thus, M. tuberculosis DNA can be detected by PCR in gastric aspirates of many children with primary tuberculosis, despite that specimens from these patients are negative by culture. Multiple samples must be tested to optimize the diagnostic yield.