Introduction: Identifying all randomized trials relevant to a review topic is a large and daunting task. Thus, clinicians and medical researchers increasingly rely on computer databases to identify trials. However, the efficiency and precision of searches using such databases is rarely assessed. The present study reports an evaluation of two widely used computerized databases of medical and psychiatric literature.

Objective: The aims were: (i) to identify all randomized trials in the International Journal of Eating Disorders; (ii) to assess the number of those that were present in two electronic databases (Psychlit and Embase); and (iii) to investigate the precision of two 100% sensitive electronic search strategies for trials in this journal.

Methods: The journal was hand-searched for randomized controlled trials. Reports of all 40 trials identified were sought in Psychlit and Embase. Database reports were inspected for methodological words and phrases, which were then combined into two search strategies capable of reidentifying all reports of trials. The precision of these searches was calculated.

Results: Psychlit contained 27 (67%) trials, and Embase 35 (88%) trials. The 100% sensitive Psychlit strategy identified 289 reports and the Embase strategy, 391. The precision of both searches was poor.

Discussion: Sensitive computer searches were imprecise. To identify one randomized trial, over 10 electronic reports had to be inspected. Significant proportions of the randomized trials were also absent from both electronic databases. Some strategies for improvements in efficiency and accuracy of database searches are proposed.