Peer Review: Types, Process, and What Reviewers Actually Do

Peer review is the process by which a manuscript, grant application, or other piece of scholarly work is evaluated by independent experts in the same field before it is accepted, funded, or published. It is the mechanism the research system relies on to catch errors, weigh significance, and put a credibility marker on work before it enters the permanent scholarly record — the difference, administratively, between a preprint or a working draft and a citable, credentialed output that counts toward a CV, a grant report, or a tenure case.

This page covers the foundational mechanics: what peer review is for, how a typical journal review cycle runs from submission to decision, what reviewers are actually asked to judge, the outcomes a review can produce, and why the process matters specifically to research administration. For a side-by-side comparison of the different identity-disclosure models (who knows whose identity, and when), see Single-anonymized vs. double-anonymized vs. open peer review — that page covers the anonymization models in detail and is not repeated here.

What peer review is for

Peer review exists to do two things at once:

• Quality control. Independent subject-matter experts check a manuscript’s methodology, reasoning, evidence, and framing before it is published — catching errors, unsupported claims, or gaps in the literature that an editor alone would likely miss.
• A credibility signal. A published, peer-reviewed article carries an implicit certification: qualified reviewers and an editor looked at this work and judged it sound enough to publish. Funders, hiring and tenure committees, other researchers deciding what to cite, and journalists deciding what to report all rely on that signal, imperfect as it is, because verifying every claim in every manuscript from scratch is not practical at scale.

That second point is exactly what predatory publishing exploits: a “journal” that skips or fakes the review step is manufacturing the appearance of that credibility signal without doing the underlying work. See How to Identify Predatory Journals and Publishers and the dictionary entry for predatory journal for how to tell the difference.

The peer review process, step by step

Exact workflows vary by publisher and by manuscript-tracking system (see Editorial Manager vs. ScholarOne vs. OJS for how the major submission platforms differ), but the sequence below is the common shape across most journals:

1. Submission. The author(s) submit the manuscript through the journal’s online system, typically along with a cover letter, any required disclosures (conflicts of interest, data availability, ethics/IRB approval), and — where the journal requires it — a CRediT contributor statement.
2. Editorial screening. A handling editor checks the manuscript is within the journal’s scope, meets basic formatting/ethical requirements, and clears a plagiarism/similarity check before it goes any further. A substantial share of submissions are rejected at this stage without ever reaching a reviewer (“desk rejection”).
3. Reviewer assignment. The editor identifies and invites independent experts in the manuscript’s specific subject area, checking for conflicts of interest (co-authorship history, institutional overlap, competing projects) before inviting them. ICMJE’s guidance places responsibility for selecting appropriate reviewers, and for ensuring they have access to everything they need to evaluate the work, on the editor.
4. Review. Reviewers evaluate the manuscript against the criteria below and submit a written report, usually with a recommendation. Most journals ask for two or more independent reviews per manuscript. Under COPE’s Ethical Guidelines for Peer Reviewers, reviewers are expected to keep the manuscript strictly confidential, declare any competing interests, decline if they lack the expertise to assess it fairly, and be timely.
5. Editorial decision. The editor weighs the reviewers’ reports — but is not bound by them. ICMJE is explicit on this point: “a peer-reviewed journal is under no obligation to follow reviewer recommendations” — the editor makes the final call and can also reject a paper the reviewers liked, or vice versa.
6. Revision (if requested). If the decision is a revision rather than an outright accept or reject, the authors respond to reviewer comments point by point and resubmit. This can go through more than one round.
7. Final decision and production. Once a version is accepted, it moves into copyediting, typesetting, and formal publication — at which point it typically receives a DOI and becomes the version of record.

What reviewers actually evaluate

A reviewer’s report is not a proofread. Reviewers are generally asked to judge:

• Originality and significance — does this add something not already established in the literature, and does it matter?
• Soundness of methodology — is the study design, statistical approach, or analytical method appropriate to the question being asked, and executed correctly?
• Validity of the evidence and conclusions — do the results actually support the claims the authors make, or do the authors overreach?
• Clarity and completeness — is there enough detail (methods, data, code) for another researcher to understand and, ideally, reproduce the work?
• Ethical compliance — proper approvals (e.g. IRB/ethics committee sign-off for human-subjects research), appropriate disclosures, and no evidence of research misconduct.
• Fit and framing — is the manuscript positioned correctly against prior work, and does it belong in this particular journal?

Reviewers are explicitly not supposed to use a manuscript’s ideas or data for their own purposes before it is published, and are expected to keep everything they see confidential — this is a recurring theme across ICMJE’s and COPE’s guidance, not an incidental courtesy.

How to conduct a peer review: a practical walkthrough

Being asked to review is different from knowing what reviewers evaluate in the abstract. A workable sequence, consistent with COPE’s Ethical Guidelines for Peer Reviewers and ICMJE’s reviewer-conduct expectations:

1. Decide whether to accept the invitation. Say yes only if the manuscript is genuinely within your expertise, you have no undisclosed competing interest (co-authorship history, institutional overlap, a rival project), and you can realistically meet the deadline. Decline promptly rather than sitting on an invitation — editors need to find another reviewer if you can’t do it.
2. Read the manuscript twice. A first pass for the overall argument, contribution, and fit; a second, close pass checking methodology, evidence, and whether the conclusions are actually supported by the results reported.
3. Evaluate against the standard criteria — originality/significance, methodological soundness, whether the evidence supports the conclusions, clarity/completeness, ethical compliance, and fit for the journal (see below).
4. Structure the report. Most journals expect: a short summary of the manuscript in your own words (confirms you understood it and gives the editor context); major comments (substantive issues that affect the paper’s validity or contribution); minor comments (smaller, more easily fixed points); and, where the platform supports it, confidential comments to the editor only — used for concerns like suspected misconduct or a conflict you want the editor aware of, not for softening feedback you’re unwilling to say to the authors directly.
5. Be specific and evidence-based. “The statistics are wrong” is not actionable; pointing to the specific test, assumption, or table that’s the problem is. See worked examples of reviewer comments and, when a report needs to be triaged, how to prioritize major vs. minor comments under time pressure.
6. Recommend a decision, but don’t over-claim authority. Your report informs the editor’s decision; per ICMJE, the editor is not obligated to follow it. Frame your recommendation as your professional judgment, not a verdict.

For the fuller picture of what it means to serve as a reviewer — including how reviewing gets recognized on ORCID — see Peer Reviewer: Role, Ethics & How-To. If you’re on the receiving end of a review instead, see How to Write a Response to Reviewers Letter and Desk Rejection: What It Means and How to Avoid It.

Common outcomes

Most journals use some version of the following decision categories, though exact labels vary:

• Accept — usually with minor, non-substantive edits (or, rarely, “accept as is”).
• Minor revision — the core work is sound; the authors need to address specific, bounded concerns (clarify a method, add a citation, tighten the discussion) before acceptance, usually without a further full re-review.
• Major revision — reviewers have identified substantive issues (methodological gaps, insufficiently supported conclusions, missing analyses) that require real additional work and, typically, another round of review once resubmitted.
• Reject — the manuscript does not meet the bar for the journal, whether because of flawed methodology, insufficient significance, poor fit, or ethical concerns. A rejection from one journal does not mean a paper can’t be published elsewhere; resubmission to a different, often more specialized or lower-tier, journal after a rejection is routine.
• Reject and resubmit — used by some journals for manuscripts that need work extensive enough that the original submission is formally closed, with an invitation to resubmit as a new submission once revised.

Types of peer review: a model-by-model comparison

Terminology in this space used to be inconsistent across publishers (“blind review” meant different things at different journals). ANSI/NISO Z39.106-2023, Standard Terminology for Peer Review — developed from an STM Association peer-review taxonomy working group and formalized as a NISO standard in July 2023 — fixed this by defining review models along four dimensions: identity transparency, who the reviewer interacts with, what review information gets published, and whether post-publication commenting is enabled. See the dictionary entry for ANSI/NISO Z39.106-2023 for the full standard. The table below applies those dimensions to the models a researcher or administrator is actually likely to encounter:

Model	Identity disclosure	What gets published	Where it’s used
Single-anonymized (formerly “single-blind”)	Reviewers know author identity; authors do not know reviewer identity	Article only — reports stay internal	The default model at most subscription and hybrid journals
Double-anonymized (formerly “double-blind”)	Neither party knows the other’s identity (author details stripped from the manuscript before review)	Article only	Common in social sciences and humanities; harder to enforce in narrow subfields where authorship is guessable from the work itself
Open peer review	Author and reviewer identities are both disclosed to each other	Varies by journal — sometimes just identities, sometimes full reports too	BMJ, BMC journals, and others that have moved away from anonymity on accountability grounds
Transparent peer review	Can be anonymized or open	Full review reports and editor decision letters are published alongside the article, usually with the authors’ response	Nature Communications’ transparent peer review option, EMBO Press journals
Collaborative peer review	Varies	Article; sometimes a single consolidated report	Reviewers consult with each other (and often the editor) before finalizing a decision, rather than submitting fully independent reports — used at Frontiers (“collaborative review”) and, in a different form, eLife’s consultative model
Post-publication peer review	Open, by definition	Commentary and critique happen in public after the article is already live	PubPeer, journal comment sections, F1000Research’s publish-then-review model — see Post-Publication Peer Review and F1000Research and Open Peer Review
Registered Reports	Varies by journal’s normal policy	Study design (Stage 1) is reviewed and provisionally accepted before data collection; results (Stage 2) are reviewed only for adherence to the approved protocol	Over 300 journals across disciplines — see Registered Report and how the two-stage process works

The three identity-disclosure models most searchers ask about — single-anonymized, double-anonymized, and open — are compared dimension by dimension, with worked examples, on Single-anonymized vs. double-anonymized vs. open peer review; that page goes deeper on those three specifically and isn’t repeated here. Community-driven open review of preprints (rather than journal submissions) is its own model again — see PREreview.

What the evidence says: studies comparing models have generally found that revealing reviewer identity to authors (open review) does not measurably change review quality or recommendation harshness, and that double-anonymization is often imperfect in practice — reviewers in a narrow subfield can frequently guess authorship from the topic, methods, or self-citation pattern even when names are stripped. No single model has been shown to eliminate bias; each trades one risk for another (anonymity can shield low-quality or careless reviews from accountability; open identity can make reviewers reluctant to give harsh-but-honest feedback to a senior colleague).

Peer review doesn’t stop at the acceptance decision, either — a growing body of practice covers commentary and correction after publication. See Post-Publication Peer Review: From Concern to Correction.

Why peer review matters for research administration specifically

For a standards body focused on research administration, peer review matters less as an editorial ritual and more as an infrastructural gate: it’s the step that converts a manuscript or dataset into a credentialed, citable output that downstream systems can rely on.

• It’s what funders and tenure/promotion committees are actually counting. Grant reports, biosketches, and tenure dossiers routinely require peer-reviewed publications specifically — not preprints, not working papers — because the review step is the field’s proxy for “this has been vetted.” That’s precisely why predatory venues that fake the process are a live compliance and reputational risk for institutions, not just an inconvenience for individual authors.
• It generates metadata that research information systems depend on. A peer-reviewed article typically acquires a DOI, publisher and journal identifiers, and (where the reviewer participates) a record in the reviewer’s ORCID record — several publishers, working through Web of Science and similar third-party recognition services, let reviewers deposit a verified peer-review credit against their ORCID iD even where the review content stays confidential. That record is exactly the kind of structured, machine-readable contribution data a CRIS (current research information system) needs to credit reviewing service, not just authorship.
• It interacts directly with authorship and contribution accounting. CASRAI’s own CRediT taxonomy is frequently disclosed alongside peer review at submission, and ICMJE’s authorship criteria and COPE’s authorship guidance are both applied, in practice, during the editorial and review stages — a paper’s authorship list is scrutinized as part of, not separately from, the review process.
• It’s the gate retractions and corrections sit downstream of. When peer review misses something, the correction mechanism (errata, expressions of concern, retraction) is a distinct, later process — see How a Retraction Actually Happens — but it exists precisely because peer review, while a real check, is not a guarantee.

Known limitations, honestly stated

Peer review is a real check, not an infallible one, and standards bodies are candid about this rather than treating it as beyond question. ICMJE’s own recommendations describe peer review as “an important extension of the scientific process,” while also noting that its actual value has been disputed within the field. Commonly cited limitations include:

• Reviewer availability, turnaround time, and reviewer fatigue. As submission volume has grown faster than the pool of willing, qualified reviewers, editors report increasing difficulty finding reviewers who will accept an invitation, and reviewers who do accept are stretched across more requests than in the past — a dynamic widely referred to as reviewer fatigue, and a direct contributor to slower and more variable turnaround times.
• Subjectivity and inconsistency between reviewers. Reviewer judgments vary — sometimes substantially — between reviewers assessing the identical manuscript; agreement between independent reviewers on accept/reject recommendations is imperfect even at well-run journals.
• Bias. Unconscious bias toward well-known authors or prestigious institutions has been documented under models where reviewer identity is disclosed to authors but not vice versa (see the model-comparison table above).
• Peer review checks plausibility and methodological soundness, not reproducibility. A peer-reviewed result can still fail to replicate — the review step evaluates whether a study was designed and reported soundly, not whether its findings will hold up under independent repetition.
• Peer review cannot reliably detect deliberate fraud. Reviewers evaluate a manuscript in good faith, on the assumption the reported data and methods are genuine; they are not auditors, and typically have no way to verify raw data, detect fabricated results, or catch a sophisticated paper-mill submission designed to look legitimate. See peer review fraud and integrity risks below for how that gap has actually been exploited, and why detection has increasingly moved to statistical and cross-publisher tools rather than resting on the reviewer’s judgment alone.

None of this makes the process worthless; it’s why the field also layers on post-publication mechanisms (open commentary, replication studies, corrections, and retraction where warranted) rather than treating the initial review as the last word.

Peer review fraud and integrity risks

Beyond ordinary limitations, peer review is also a target for deliberate manipulation. This is a research-integrity issue in its own right, not just an editorial-quality one, and it has grown alongside submission volume:

• Fake reviewer rings and reviewer-suggestion exploitation. Many manuscript-tracking systems let authors suggest potential reviewers — a convenience that has been directly exploited. In a documented 2014 case, SAGE Publications retracted 60 articles from the Journal of Vibration and Control after a 14-month investigation found a single author had created roughly 130 fabricated or aliased reviewer email accounts on the submission platform, then used them to review his own and colleagues’ papers under fake identities. It is one of the clearest publicly documented examples of a “peer review ring,” and the underlying technique — supplying reviewers who are fabricated or complicit rather than genuinely independent — remains a known paper-mill tactic.
• Compromised or exploited editor roles. Special/guest-issue editor positions carry lighter vetting at many journals than a permanent editorial board seat, which paper mills have exploited to wave large batches of fabricated manuscripts through with minimal real review. A 2023 Science/AAAS investigation documented paper mills bribing or otherwise compromising editors at reputable journals, including via guest editor roles. See Guest Editor Vetting for how publishers now screen these appointments.
• Paper mills at scale. A paper mill sells fabricated manuscripts, authorship slots, or manipulated peer-review outcomes as a commercial service. The largest publicly documented case is Hindawi (acquired by Wiley in 2021): following a wave of paper-mill infiltration, Wiley paused special issues in 2023, and cumulative retractions across former-Hindawi titles passed 11,000 articles by 2024, ultimately leading Wiley to retire the Hindawi brand entirely. Publishers now coordinate on detection through the STM Integrity Hub and the United2Act coalition. See the dictionary entry for paper mill and, for the fuller case history, the Hindawi/Wiley timeline.
• AI-generated reviews. A reviewer pasting a confidential manuscript into a generative-AI tool to draft their report raises two distinct problems: a confidentiality breach (the manuscript’s unpublished content is exposed to a third-party system), and a quality problem (an AI-drafted review may misstate what the manuscript actually says). Publisher policies differ in specifics but converge on the core point — several major publishers now explicitly bar reviewers from generating review text with tools like ChatGPT, while permitting narrower, disclosed assistive uses. See AI in Peer Review and the publisher-by-publisher policy comparison.

None of this means peer review “doesn’t work” — most manuscripts are reviewed by genuine, independent experts, and the cases above became public specifically because detection mechanisms (statistical anomaly detection, cross-publisher data sharing, whistleblowers, and post-publication scrutiny) caught them. But it does mean peer review alone cannot be treated as a guarantee against fraud, which is exactly why standards bodies now layer additional checks — image-integrity screening, data-availability requirements, and cross-publisher integrity tools — on top of the traditional review step rather than relying on it alone.

Common questions

What is peer review in research?

It’s the evaluation of a manuscript, grant proposal, or other scholarly output by independent experts in the same field, before that work is published or funded, to check its methodology, evidence, and significance and to certify it meets the field’s basic standards.

Why is peer review important?

It’s the closest thing scholarly publishing has to independent quality control before something enters the permanent record, and it functions as the credibility signal that funders, employers, other researchers, and the public rely on to distinguish vetted research from an unreviewed claim. That’s true whether the context is a research article, a psychology study, a science paper generally, or student coursework framed as “peer review” practice — the underlying function is the same independent-check role, even though a classroom exercise obviously carries far lower stakes than journal review.

What are the different types of peer review?

Seven models are in active use: single-anonymized, double-anonymized, open, transparent, collaborative, post-publication, and Registered Reports — compared in the table above. The most commonly searched distinction is the identity-disclosure model (single-anonymized vs. double-anonymized vs. open), covered dimension-by-dimension on this site’s comparison page.

What is open peer review?

Open peer review is any model where reviewer and author identities are disclosed to each other, rather than kept anonymous. Some journals extend “open” further to mean the review reports themselves are published alongside the article (closer to what this page calls transparent review) — check a specific journal’s policy, since “open peer review” is used slightly inconsistently across publishers. See the dictionary entry for open peer review.

How do I conduct a peer review?

At minimum: confirm you have the expertise and no conflict of interest, read the manuscript closely, evaluate it against the journal’s stated criteria (originality, methodology, evidence, clarity, ethics, fit), and write a structured report with a summary plus specific, evidence-based major and minor comments. See the step-by-step walkthrough above.

How do I know if an article is peer reviewed?

Check the journal’s own “About” or “Peer review policy” page, which should state its review process explicitly. Most academic library databases (and Google Scholar’s advanced search, and dedicated databases like PubMed or Scopus) let you filter search results to peer-reviewed journals directly. A journal’s inclusion in a recognized index (e.g. Web of Science, Scopus, or a discipline-specific indexing service) is a reasonable proxy, though not a guarantee, since indexing criteria and review rigor aren’t identical things.

Where can I find peer-reviewed articles?

University and institutional library databases (which typically let you filter to peer-reviewed content specifically), PubMed/MEDLINE for biomedical topics, Scopus and Web of Science across disciplines, and Google Scholar all index peer-reviewed literature, alongside preprints and other non-reviewed material that isn’t always clearly distinguished in a basic search — check the source or use a database’s peer-review filter rather than assuming every result is reviewed.

How long does peer review typically take?

There’s no universal figure — it depends heavily on field, journal, and reviewer availability, and can range from a few weeks to many months, especially if a manuscript goes through multiple revision rounds. Journals increasingly publish their own historical median turnaround times; check the specific journal’s website rather than relying on a field-wide average, which this page won’t fabricate a number for.

Is peer review the same as editing?

No. Editing (copyediting, proofreading) checks language, formatting, and style, generally after a manuscript has already been accepted. Peer review evaluates the substance — methodology, evidence, significance — before an acceptance decision is made, and is normally performed by subject-matter experts in the field rather than editorial staff.

This page is a foundational overview of peer review. For the anonymization/identity-disclosure models specifically, see the comparison page. For spotting venues that fake the process, see How to Identify Predatory Journals and Publishers. For more on CASRAI’s broader publishing-integrity work, see the Scholarly Publishing cluster.