The accuracy of five imaging modalities for the diagnosis of lumbar herniated nucleus pulposus (HNP) is compared prospectively in 124 patients, all of whom underwent surgical exploration. All tests were read independently of each other and the level of confidence in each diagnosis was recorded. The results are based on negative (106) as well as positive (125) findings at the 231 disc sites (level and side) explored. Computed tomography-discography (disco-CT) was the most accurate test (87%) compared to 77% for CT-myelography (myelo-CT), 74% for CT, 70% for myelography, 64% for disc injection pain, and 58% for discography. The false positive rate was lower for disco- CT (19%) than for myelo-CT (24%), CT (24%), and myelography (30%). The false negative rate was also lower for disco-CT (8%) than for myelo-CT (22%), CT (29%), and myelography (30%). Disco-CT was the most accurate test (94%) in patients who had prior disc surgery compared with 81% for myelo-CT, 80% for CT, and 74% for myelography. Disco-CT was also the most accurate test for patients with foraminal HNP (91% compared with 71% for CT, 65% for myelo-CT, and 58% for myelography). Disc injection reproduced the patient's clinical pain pattern in only 36% of herniated discs. This test has high specificity (89%), but low sensitivity (43%). The risks from myelography followed by discography within a 72-hour period are similar to those reported for myelography alone. Disco- CT is the most accurate of these tests (P<0.05) for the diagnosis of lumbar HNP. It has both high sensitivity and specificity and should be considered for patients with suspected disc herniation whose other tests are nondiagnostic, especially those with possible foraminal or recurrent HNP.