Objective To study the methodology of Chinese literature retrieval. Methods The manual review of the literature was served as the “gold standard” against database search strategies (the diagnostic tests). We selected original articles about treatment, rehabilitation and randomized controlled trials from 31 journals. The articles were downloaded from the Chinese Biomedical Database (CBM). We selected potentially useful words through a word frequency analysis and determined the frequency of all the words in the titles, abstracts, and subject indexes. All the selected journals functioned as a closed database. The sensitivity, specificity and precision of all the high frequency words were calculated and the high frequency words of large sensitivity×precision were considered as final searching words. All the searching strategies were produced by computer programe which consisted of all searching words, title field and abstract field. Meanwhile, the sensitivity, specificity, precision and NNR (number needed to read) were calculated. Among the strategies, those comprised of all searching words would be used in CBM disc database, those of title and abstract words in Chinese Web Databases. The best strategies were those of high sensitivity and high specificity. Results 2 570 articles were selected and 45 articles met the gold standard. The strategies emphasized sensitivity were “therapeutic use OR random OR control(for CBM disc) and multicenter OR therapeutic outcome OR random (for Chinese online databases)”. The strategies emphasized specificity were “placebo OR prospective study(MH) OR double blind OR random controled trial (MH) (for CBM disc) and placebo OR prospective) OR double blind OR efficiency (for Chinese online databases).”Conclusions The method is optimal for Chinese literature databases