How does peer review work? In short: a journal editor receives a manuscript, sends it to two or more experts in the same field, reads their reports, and then decides whether the paper is accepted, revised, or rejected. The experts are volunteers. Their advice shapes what science gets published, but the final call always belongs to the editor.
The word itself carries the whole idea. A peer, as the Cambridge Dictionary puts it, is an equal in age, education or social class. Reviewers are not examiners standing above the author. They are colleagues reading a colleague's work, which is why the process is called peer review and not expert review.
It is also worth saying what peer review is not. It is not a stamp of truth. It is a quality filter run by busy people, and it misses things. Understanding how the filter is built — who gets chosen, what the anonymity models hide, and how decisions get made — helps any reader judge how much weight a published paper deserves.
How does a journal choose its reviewers?
The editor starts with the manuscript and a question: who knows this exact corner of the field well enough to judge it, and has no obvious conflict with the authors? Most journals keep a pool of past reviewers, search recent citation records, and sometimes ask the authors themselves to suggest names. Authors are usually also invited to rule people out, which is meant to block rivals, not friends.
Selection is a balancing act. A reviewer too close to the authors may go easy. A competitor may go hard. A reviewer from a different subfield may miss the technical point entirely. Editors try to combine at least one specialist who knows the method inside out with someone who can judge whether the bigger claim matters. Turnaround pressure shapes everything; the best-qualified people are often the busiest, so the second-best choice sometimes wins.
Reviewers are unpaid. They review because the system runs on reciprocity — you read for others, and others read for you. That bargain is the quiet engine of the whole enterprise, and it is also its weak point: when the pool of willing reviewers thins, quality and speed both suffer.
What do the anonymity models actually hide?
Three models dominate, and each hides something different. In single-blind review, the most common model, the reviewer knows who wrote the paper but the author never learns who reviewed it. The reviewer can be frank without social cost. The cost is that the reviewer's own biases — about the authors, their institution, or their prior work — go unexamined.
Double-blind review hides both directions: the reviewer does not know the authors, and the authors do not know the reviewer. In theory this strips status and reputation out of the judgement. In practice it leaks. Authors describe their own methods so specifically that they often identify themselves, and the paper's reference list can give the game away. We have covered How Double-Blind Peer Review Actually Works, and Where It Fails in detail elsewhere.
Open review removes the mask entirely. Reviewer names may appear on the published report, or the reports themselves may be published alongside the paper. Supporters say transparency improves conduct; critics say anonymity is what lets a junior reviewer contradict a famous one. Neither side has a knockout argument, which is why journals split on the question.
What do reviewers actually check?
A reviewer reads for several things at once. Is the question worth asking? Is the method capable of answering it? Do the data actually support the claims as written? Is the work original enough, and is it presented honestly — limitations stated, prior work credited, nothing oversold? Most reports also flag smaller problems: a statistic reported wrongly, a figure that does not match its caption, a key reference missing.
Reviewers do not re-run the experiments. They do not check the raw data, unless the journal asks for it. They judge what is on the page. This is the single most misunderstood fact about peer review: it is a reading job, not an audit. Fraud that is well disguised on the page can pass, which is one reason retractions happen after publication rather than before. Our Retraction Watch: Why Papers Get Pulled and What It Teaches piece follows what happens then.
The report itself usually has two layers: comments to the editor, which may be frank about the authors, and comments to the authors, which are meant to be usable. A good report says what would fix the paper, not just what is wrong with it.
How does an editor turn reports into a decision?
Once the reports arrive, the editor weighs them like a judge weighing testimony. Full agreement is easy. Full disagreement is common, and then the editor must decide which objection is fatal and which is taste. The standard paths look like this:
- Accept — rare on a first pass; the paper is publishable as it stands.
- Minor revision — the science holds; the fixes are mechanical. Often sent back only to the editor, not the reviewers.
- Major revision — the paper is worth saving but needs new analysis, new experiments, or serious rewriting. Revised versions usually go back to the original reviewers.
- Reject and resubmit — a new paper is invited once the problems are fixed.
- Reject — the problems are beyond repair at this journal, or the paper fits better elsewhere.
When reviewers deadlock, editors have options: seek one more opinion, overrule one report with a reason, or reject as the safe course. Overruling is more common than outsiders assume. The editor, not the reviewers, owns the decision — and the authors' appeal, where a journal allows one, goes to the editor too, sometimes to a different editor as a second look.
Timing varies wildly by field and journal, from a few weeks to the better part of a year, and a major-revision loop can add another full round. Authors planning a career around a publication date learn to budget for this.
What peer review cannot do
Peer review improves papers; it does not certify them. It catches most sloppy reasoning and much unclear writing. It cannot detect a fabricated dataset that looks plausible, and it cannot judge findings that arrive before the field knows how to interpret them. The system's defenders describe it accurately as the best filter anyone has built so far — a guarantee of process, not of product. The Cambridge Dictionary's corpus of usage even captures the modest framing: the phrase that appears in real acknowledgments is thanks for "valuable insights and constructive comments," not certification.
That is why the surrounding machinery matters. Some journals now trial Registered Reports: Peer Review Before the Data Exist, which reviews the design before results are known, removing the temptation to shape findings after the fact. Others are writing rules for whether reviewers may use AI at all. And authors choosing where to submit should learn the warning signs of journals that fake the whole process, because the words "peer reviewed" on a website guarantee nothing by themselves.
Our analysis: what this means for readers of research
For anyone outside the system, three takeaways hold. First, peer review is a filter, not a verdict — a published paper has cleared a bar, not proven its case. Second, the quality of that bar depends on who was holding it, and reviewers are chosen by one editor with imperfect information. Third, the process is human end to end: volunteers, deadlines, disagreements, and a final judgement call. Read the limitations section the authors wrote, not just the abstract, and you are already reading more critically than the system requires of anyone. For more on the machinery of publishing, see our Journals section.
