Scientific journals can promote ethical publication practices through policies on conflicts of interest. However, the prevalence of conflict of interest policies and the definition of conflict of interest appear to vary across scientific disciplines. This survey of high-impact, peer-reviewed journals in 12 different scientific disciplines was conducted to assess these variations. The survey identified published conflict of interest policies in 28 of 84 journals (33%). However, when representatives of 49 of the 84 journals (58%) completed a Web-based survey about journal conflict of interest policies, 39 (80%) reported having such a policy. Frequency of policies (including those not published) varied by discipline, from 100% among general medical journals to none among physics journals. Financial interests were most frequently addressed with relation to authors; policies for reviewers most often addressed non-financial conflicts. Twenty-two of the 39 journals with policies (56%) had policies about editors' conflicts. The highest impact journals in each category were most likely to have a published policy, and the frequency of policies fell linearly with rank; for example, policies were published by 58% of journals ranked 1 in their category, 42% of journals ranked third, and 8% of journals ranked seventh (test for trend, p = 0.003). Having a conflict of interest policy was also associated with a self-reported history of problems with conflict of interest. The prevalence of published conflict of interest policies was higher than that reported in a 1997 study, an increase that might be attributable to heightened awareness of conflict of interest issues. However, many of the journals with policies do not make them readily available and many of those policies that were available lacked clear definitions of conflict of interest or details about how disclosures would be managed during peer review and publication.