Recognition of the need for more scientifically acceptable review strategies is increasing. Systematic reviews have rarely been performed on randomized controlled trials (RCTs) in the field of epilepsy. We have 355 published RCTs on a database, which will be used as a data source from systematic reviews. We wished to test the sensitivity and precision of two Medline search strategies, and to identify reasons why RCTs were missed. Both a basic and a comprehensive search strategy were used to search Medline from 1966 to 1993. Three journals (Epilepsia, Epilepsy Research, Acta Neurological Scandinavica) were searched by hand. The sensitivity and precision of these Medline search strategies in these three journals was calculated. We investigated articles missed by our comprehensive search to ascertain why they had been missed. Of the 308 RCTs on the database published between 1966 and 1993, 275 were found by our comprehensive Medline search, whereas 103 were found by our basic search. The overall precision of our comprehensive and basic searches was 35 and 72%, respectively. The overall sensitivity of our comprehensive and basic searches was 86 and 66%, respectively. Our comprehensive search failed to find 16 RCTs published in the three journals searched by hand. Articles were missed because they were inadequately indexed on Medline and/or did not contain adequate methodological information in their titles or abstracts. To find all relevant trials, a search strategy must be as sensitive as possible, but this sensitivity inevitably involves loss of precision. The National Library of Medicine is addressing indexing problems with Medline, but the quality of reports also must be improved to enhance retrieval.