The existence of a Register of Controlled Trials in Perinatal Medicine (National Perinatal Epidemiology Unit, Oxford, England) has offered an opportunity to assess the efficacy of online searching of MEDLINE, an example of a broad bibliographic database. Retrieval of all relevant randomized control trials (RCTs) in a given field is important in analyses in which results are pooled (meta-analyses). Reports of RCTs of prevention and treatment of neonatal hyperbilirubinemia and prevention of intraventricular hemorrhage (IVH) for the years 1966-1983 were sought in both the Register and MEDLINE files. Comparison of subject searches revealed a number of unlisted papers in each file that were then found to be present by an author search. In the MEDLINE searching an amateur was clearly less efficient than an expert, but the expert recovered only 29% of the relevant hyperbilirubinemia papers available in MEDLINE, and only 56% of the identified IVH RCTs. Some of the deficiencies in recovery have been corrected by indexing improvements, such as the capability of identifying text words in abstracts, and the addition of new medical subject heading terms (MeSH) such as RANDOM ALLOCATION. Efficiency will be best facilitated by authors and editors keeping the MeSH terms used by MEDLINE indexers in mind when they compose titles and abstracts.