We apply a novel mistake index to assess trends in the proportion of corrections published between 1993 and 2014 in Nature, Science and PNAS. The index revealed a progressive increase in the proportion of corrections published in these three highquality journals. The index appears to be independent of the journal impact factor or the number of items published, as suggested by a comparative analyses among 16 top scientific journals of different impact factors and disciplines. A more detailed analysis suggests that the trend in the time-to-correction increased significantly over time and also differed among journals (Nature 233 days; Science 136 days; PNAS 232 days). A detailed review of 1,428 errors showed that 60% of corrections were related to figures, authors, references or results. According to the three categories established, 34.7% of the corrections were considered mild, 47.7% moderate and 17.6% severe, also differing among journals. Errors occurring during the printing process were responsible for 5% of corrections in Nature, 3% in Science and 18% in PNAS. The measurement of the temporal trends in the quality of scientific manuscripts can assist editors and reviewers in identifying the most common mistakes, increasing the rigor of peer-review and improving the quality of published scientific manuscripts.