This study aims to compare handsearching to a basic MEDLINE search for the identification of reports of randomized trials in specialized health care journals. Twenty-two specialized health care journals, published in the U.K., were handsearched for all reports of controlled trials (as defined by the Cochrane Collaboration). The reports of trials, which were judged to be definitely randomized, were identified from a random sample of three years per journal and form one element of this study. A MEDLINE search using the publication type terms 'randomized controlled trial' and 'controlled clinical trial' was also performed for the same journal years. The reports of trials retrieved by handsearching were then compared against those retrieved from the MEDLINE search, to identify differences in retrieval between the two techniques. Reports of randomized trials identified by the MEDLINE search but not found by handsearching were individually assessed to see if they met the Cochrane eligibility criteria for a report of a randomized trial. A total of 714 reports of randomized trials were found by using a combination of both handsearching and MEDLINE searching. Of these, 369 (52 per cent) were identified only by handsearching and 32 (4 per cent) were identified only by MEDLINE searching. Of the reports identified only by handsearching, 252 had no MEDLINE record, of which 232 (92 per cent) were meeting abstracts or published in supplements; 117 (25 per cent) of the 462 reports of randomized trials which had a MEDLINE record were missed by the electronic search because they did not have either of the publication type terms 'randomized controlled trial' or 'controlled clinical trial'. This proportion varied depending on when the reports of randomized trials were published (that is, before or after the introduction of the MEDLINE publication type terms above). The highest additional yield from handsearching compared to MEDLINE searching was for reports of randomized trials published prior to 1991 and from handsearching the non-MEDLINE indexed parts of a journal. The results of this study suggest that a combination of MEDLINE and handsearching is required to identify adequately reports of randomized trials.