Methods: We searched Issue 1, 2007 of The Cochrane Library to identify CSRs of Traditional Chinese Medicine (TCM). We telephoned the original authors of studies included in the reviews to discuss their methods of randomisation. Results: Twenty-nine CSRs met our inclusion criteria. Three reviews included 40, 66 and 75 studies, respectively; three were empty reviews, and two reviews included only one study; the rest included 3 to 23 studies. All of the 352 studies included in the 29 reviews were published in China. Seven reviews including a total of 155 studies were selected for study. By interviewing the study authors by telephone, we were able to contact 108 of the 155 authors of studies included in those reviews. Four authors refused to answer our questions. Forty-six authors (42.6%, 46/108) told us that they have had previously developed a research protocol. Of these 46 studies, only 12 could be authenticated as true RCTs (11.1%, 12/108). The authors of the other 96 studies sincerely told us the randomization methods they used were not real RCTs, of them, (1) 88 studies had not generated an allocation sequence before recruiting the patients, and they optionally allocated the patients; (2) two studies did not allocate the patients according to the allocation sequence; (3) six studies were retrospective case reports indeed. In consequence, they were confirmed not to be RCTs. In four Cochrane reviews, both the components of the formulations used in the included studies and the herb names in different languages were presented in additional tables while in another four Cochrane reviews, only components of the formulations were listed. Conclusions: Some CSRs of TCM have a potential risk of bias. We suggest that Cochrane review authors check with the original study authors to verify that studies claiming to be RCTs are indeed randomized. For this, we suggest to check the following key factors: 1. Was a protocol developed before the study started? 2. Was an allocation sequence generated before recruiting participants, and how?