Introduction: A crucial requirement of any Cochrane Review Group is to identify as many relevant trials as possible. More information is required on what the best methods of trial identification are and what resources are needed.

Objective: To determine the contribution of various search strategies to a database of controlled trials, and to measure the resources required to develop and maintain such a database.

Methods: Six different methods were used to identify references to trials. We analysed the number of trials first identified by each method, and the number of trials that would have been identified by using a restricted number of strategies. We also measured the resources required.

Results: The database as of 1 April 1995 included 1089 references to 812 trials: 105 were planned or ongoing, only 19 were completed unpublished trials, and 101 trials were only published as abstracts. Of these trials: 139 (17%) were identified from Medline; 267 (33%) from hand searches of 26 specialist journals and 43 conference proceedings; 221 (27%) from searching references lists; 163 (20%) from personal contact with trialists or drug companies; 12 (1%) from dissertations; and 10 (1%) from prospective searching of Current Contents. Medline searching alone would have identified 45% of trials, hand searching alone 40%, and the combination of handsearching and Medline searching 66%. In total, 6500 references have been reviewed, 2000 of which had to be retrieved in full text. The database has taken 18 months to develop with part-time input from 3 staff and 12 volunteer hand searchers.

Discussion: Multiple search strategies need to be used to maximise the identification of relevant trials: electronic searches and hand searching alone are not sufficient. Significant resources are required.