Peer review is often described as the gold standard of scientific publishing. It is more accurately described as a filter: a structured way for editors to get independent feedback before publishing a manuscript. Like any filter, it catches some problems and misses others.
Understanding what peer review can and cannot do makes 'peer-reviewed' a more useful signal — and reduces the disappointment when peer-reviewed work later turns out to be wrong.
What peer review is
After a researcher submits a manuscript to a journal, an editor decides whether to send it out for review. Reviewers — typically two to four — are scientists with relevant expertise. They read the manuscript and write critiques. The editor weighs the critiques and decides to accept, request revisions, or reject.
Most journals use single- or double-blind review. Open peer review, where the reviewers' names are published with the paper, is growing but still a minority practice.
What peer review catches
Reviewers are good at identifying methodological problems, missing controls, weak statistical reporting, and overreach in the discussion section. They often suggest additional analyses and require the authors to address known limitations.
When peer review works well, the published paper is meaningfully clearer and more defensible than the submitted manuscript.
What peer review does not catch
Peer review does not, in most cases, detect fraud. Reviewers see the analyses authors choose to report, not the raw data. It also does not guarantee replication: a paper can be methodologically sound and still describe a finding that does not hold up when other groups try to reproduce it.
Peer review is a starting point for credibility, not a finishing line.
Preprints and post-publication review
Preprint servers — arXiv, bioRxiv, medRxiv, SSRN — publish manuscripts before peer review. Preprints accelerate the spread of research and allow open critique, but readers should treat them as preliminary.
Post-publication review on platforms like PubPeer can catch issues — image duplication, statistical errors — long after a paper appears in a journal.