To minimise the effect of reference bias in literature retrieval, it is important to use computerised search strategies that give a high yield of relevant reports. In a MEDLINE search that included the Medical Subject Heading (MeSH) "Comparative study," the recall of double-blind trials of NSAIDs in rheumatoid arthritis was 93.1% (122/131) and the precision was 19.0% (122/641). When "Double-blind method" was used, either as MeSH or text words, the recall was only 72.5% (95/131) with a precision of 22.7% (95/419). A combined search strategy increased the recall to 97.7% (128/131) with a precision of 17.3% (128/738). With the MeSH term "Random allocation" only eight relevant reports were retrieved, and none was new. By using "Clinical trials" alone, we would have missed eleven reports. We conclude that "Comparative study" is preferable to "Double-blind method" when searching double-blind trials on MEDLINE.