Randomized clinical trials (RCTs) provide the most reliable therapeutic information available. Unfortunately, there are no systemic listing of RCTs. We compared a MEDLARS search of 3686 biomedical journals for RCTs with a manual search of the medical literature for the period 1966-1982. For the former search we used subject headings (1) liver disease or (2) biliary tract disease and subheadings (1) drug therapy, (2) surgery, (3) radiotherapy, or (4) therapy, and check tags (1) comparative study or (2) clinical research. For the manual search, the contents of 34 arbitrarily selected, gastroenterologic, hepatologic, surgical, or general journals were perused. The MEDLARS search identified 160 RCTs and 29 others were found in the references of the 160. One hundred fifty-four RCTs were identified by both methods. The manual search identified 208 RCTs and an additional 34 were found in the references of the 208. The MEDLARS search identified only 107 of 208 RCTs found in the references of the 208. The MEDLARS search identified only 107 of 208 RCTs found manually in the 36 journals, an efficiency rate of 51%. We estimate that 330 hepatobiliary RCTs had been published during this 17-year period. Sixty percent of the RCTs found by MEDLARS used the key word "randomized," "double blind," or "controlled" in the title, compared to 36% in those found by the manual search. In order to retrieve RCTs, it is essential that editors require that RCTs be identified in their titles or key words by specific terms such as "controlled," "randomized," and "double blind," that papers be so catalogued and indexed, and that searchers be instructed in appropriate search strategies.