We conducted a systematic review and meta-analysis of anonymous surveys asking scientists whether they ever committed various forms of plagiarism. From May to December 2011 we searched 35 bibliographic databases, five grey literature databases and hand searched nine journals for potentially relevant studies. We included surveys that asked scientists if, in a given recall period, they had committed or knew of a colleague who committed plagiarism, and from each survey extracted the proportion of those who reported at least one case. Studies that focused on academic (i.e. student) plagiarism were excluded. Literature searches returned 12,460 titles from which 17 relevant survey studies were identified. Meta-analysis of studies reporting committed (N = 7) and witnessed (N = 11) plagiarism yielded a pooled estimate of, respectively, 1.7 % (95 % CI 1.2–2.4) and 30 % (95 % CI 17–46). Basic methodological factors, including sample size, year of survey, delivery method and whether survey questions were explicit rather than indirect made a significant difference on survey results. Even after controlling for these methodological factors, between-study differences in admission rates were significantly above those expected by sampling error alone and remained largely unexplained. Despite several limitations of the data and of this meta-analysis, we draw three robust conclusions: (1) The rate at which scientists report knowing a colleague who committed plagiarism is higher than for data fabrication and falsification; (2) The rate at which scientists report knowing a colleague who committed plagiarism is correlated to that of fabrication and falsification; (3) The rate at which scientists admit having committed either form of misconduct (i.e. fabrication, falsification and plagiarism) in surveys has declined over time.