Plagiarism is not a single act with a single definition — it is a spectrum running from verbatim copying to subtly uncredited paraphrase to reuse of an author’s own prior work. The CASRAI Plagiarism entry gives the core operational definition (appropriation of another person’s ideas, processes, results, or words without giving appropriate credit); this guide lays out the recognized types within that definition, distinguishes them from adjacent-but-different integrity issues, and links each to the standards and tools that govern how institutions and publishers handle them.
Where the “types of plagiarism” list comes from
There is no single regulatory taxonomy of plagiarism types the way there is, for instance, a federal definition of research misconduct. The commonly cited list of seven types — direct, mosaic/patchwork, paraphrasing, self-plagiarism, source-based, accidental, and complete plagiarism — is a practitioner and academic-writing-center framework popularized by plagiarism-checker vendors and writing-instruction resources (Grammarly, academic writing guides, university libguides all converge on close variants of it), not a list issued by a body like the Office of Research Integrity (ORI), the Committee on Publication Ethics (COPE), or the International Committee of Medical Journal Editors (ICMJE).
Those standards bodies instead work from a narrower, operational split. ORI’s own guidance divides plagiarism into two categories: plagiarism of text (copying a portion of text from another source without giving appropriate credit — verbatim copying, mosaic plagiarism, patchwriting) and plagiarism of ideas (thoroughly paraphrased material that still goes uncredited). That two-way split matters practically: similarity-checking software such as Turnitin, iThenticate, or Crossref Similarity Check can catch plagiarism of text because it works by string-matching against a reference corpus, but it structurally cannot catch plagiarism of ideas, since no matching string exists once the wording has been fully reworded. See CASRAI’s guide to anti-plagiarism software for how that detection actually works.
In the United States, the operative legal definition sits in federal regulation: 42 CFR Part 93, the Office of Research Integrity’s research-misconduct regulation applied to PHS-funded research, defines plagiarism as “the appropriation of another person’s ideas, processes, results, or words without giving appropriate credit.” That single-sentence definition covers every type described below — the “types” are ways the underlying appropriation happens, not separate legal categories. See the ORI dictionary entry for the fuller regulatory context.
The seven commonly cited types of plagiarism
1. Direct (verbatim) plagiarism
Copying another person’s text word-for-word, without quotation marks or citation, and presenting it as original. This is the most straightforward type to detect with similarity-checking software, because it produces an exact or near-exact string match. CASRAI’s Textual Plagiarism entry covers this and the related mosaic form in more detail, including ORI’s guideline that verbatim text must be both quoted and cited — citation alone is not sufficient once wording is copied exactly.
2. Mosaic (patchwork) plagiarism
Also called patchwriting: stitching together phrases and sentences lifted from one or more sources, with light word-swapping or minor rephrasing, into what reads as a new passage. Because the underlying wording is still substantially the source author’s, ORI classifies this under plagiarism of text alongside verbatim copying — it is closer to direct plagiarism than to genuine paraphrase, and similarity checkers frequently flag it as fragmented matches rather than one long match. Covered in depth on the Textual Plagiarism page.
3. Paraphrasing plagiarism (plagiarism of ideas)
Restating another person’s argument, findings, or structure thoroughly enough in one’s own words that no string match survives, but without attribution. This is ORI’s “plagiarism of ideas” category, and it is the hardest type to catch mechanically — a properly reworded paraphrase produces no similarity-index hit, which is precisely why Originality Reports and similarity percentages are informative signals, not proof of an absence of plagiarism. The distinguishing test is whether the source of the idea is credited, not whether the wording matches.
4. Self-plagiarism (text recycling)
Reusing one’s own previously published words, data, or findings and presenting them as new, without disclosure. Self-plagiarism is a genuinely distinct case from the other six types because it does not involve appropriating another person’s work at all — and that distinction has real regulatory weight. ORI’s own guidance states directly that 42 CFR Part 93 does not treat self-plagiarism as research misconduct, since the federal definition is scoped to “another person’s” ideas or words. COPE’s Discussion Document on text recycling nonetheless treats undisclosed self-plagiarism as a publication-ethics violation independent of that misconduct bar, because it misrepresents novelty to readers, reviewers, and journals. See the full Self-plagiarism entry for the legal-status distinction and how it differs from permissible text recycling (e.g., a disclosed methods section reused across related papers).
5. Source-based plagiarism
Citing sources in a way that misrepresents them — for example, citing a secondary source’s characterization of a study while presenting it as though the primary source had been consulted directly, citing a source that doesn’t actually support the claim, or fabricating a citation to a source that doesn’t exist. CASRAI’s Fake Citation entry covers the fabricated-source end of this category specifically.
6. Accidental plagiarism
Uncredited use of another’s words or ideas that results from carelessness — forgetting to close a quotation, losing track of which notes were paraphrase versus direct copy, or genuinely not knowing a citation was required — rather than intent to deceive. Accidental plagiarism is still plagiarism under most institutional and publisher definitions, because standard definitions (including 42 CFR Part 93’s) turn on the act of uncredited appropriation, not on demonstrated intent; institutions vary in how they weigh intent when determining consequences, but “I didn’t mean to” does not itself make a passage not count as plagiarized.
7. Complete plagiarism
Submitting another person’s entire work — a full paper, thesis, or dataset — as one’s own. This is the most severe instance of direct plagiarism rather than a mechanically distinct type, but it is commonly listed separately because institutional and publisher responses to it (typically the most serious available sanction) differ sharply from responses to a plagiarized paragraph within an otherwise original work.
Beyond text: non-textual and data-related plagiarism
The seven-type list above is built around written text, but the same underlying principle — appropriation without credit — applies to non-textual research outputs:
- Data plagiarism: presenting another researcher’s dataset, or results derived from it, as one’s own original data collection or analysis, without attribution.
- Image and figure plagiarism: reusing another author’s figures, photographs, or illustrations without permission or credit. This is a distinct problem from image manipulation, which concerns altering an image (one’s own or another’s) in a way that misrepresents the underlying data — image manipulation is a falsification concern; unauthorized reuse of someone else’s image without credit is the plagiarism concern. The two can co-occur but are evaluated separately.
- Code and software plagiarism: reusing another party’s source code or algorithms without attribution or license compliance, increasingly relevant as code and software outputs are treated as citable research products in their own right.
Adjacent misconduct categories that are often confused with plagiarism types
Several practices frequently discussed alongside “types of plagiarism” are not, technically, forms of plagiarism at all, because they don’t involve appropriating someone else’s ideas, words, or credit. Distinguishing them matters for anyone writing an institutional policy or a misconduct allegation:
- Salami slicing (splitting one dataset or study into the minimum publishable unit across multiple papers) and duplicate publication raise redundant-publication and self-citation concerns, not appropriation of another’s work — they overlap conceptually with self-plagiarism but are evaluated under separate editorial policies.
- Ghost authorship (an uncredited contributor who did substantive work) and gift/guest authorship (a credited author who did not meet authorship criteria) are authorship-misrepresentation issues, not plagiarism — no text or idea is appropriated; instead, credit is misallocated in the other direction. See CASRAI’s ghost authorship vs. gift authorship comparison for how the two differ from each other and from plagiarism.
- Research misconduct more broadly, under the U.S. federal FFP framework, includes fabrication and falsification alongside plagiarism as three legally distinct categories — fabrication and falsification concern invented or altered data, not uncredited appropriation of someone else’s work. CASRAI’s Research Integrity Fundamentals guide covers all three FFP categories together.
Plagiarism taxonomies vary by venue: the IEEE example
Publishers and institutions sometimes formalize their own operational taxonomy layered on top of the general types above. IEEE, for instance, grades plagiarism findings across five severity levels (Level One through Level Five) based on how much of a work was copied, whether it was credited at all, and whether it recurs across an author’s publication history — from an uncredited verbatim copy of an entire paper (Level One) down to verbatim copying that is credited and cited but not clearly marked as a direct quote (Level Five). Corrective actions scale accordingly, from a required apology letter up to a submission ban. See CASRAI’s IEEE Plagiarism Policy and Prohibited Authors List entry for the full scale. This is a useful illustration that “how many types of plagiarism are there” doesn’t have one universal answer — the count and the boundaries depend on which body’s framework is being applied.
Detecting the different types
Similarity-checking tools (Turnitin, iThenticate, Crossref Similarity Check) generate an Originality Report comparing a submitted document against a reference corpus of web content, journals, and previously submitted student work, and produce a similarity percentage. No vendor, COPE, or 42 CFR Part 93 sets a numeric threshold above which a similarity score constitutes misconduct — any specific cutoff in circulation is a local institutional or journal policy, not a field-wide standard (see Acceptable Similarity Percentage). Critically, this detection method is effective against the plagiarism-of-text types above (direct, mosaic, complete, and much source-based plagiarism) and structurally blind to plagiarism of ideas (thorough paraphrasing) — Turnitin’s own guidance states plainly that the similarity tool does not determine whether plagiarism has occurred, it supports a reviewer’s own judgment. On AI-assisted paraphrasing specifically, see CASRAI’s guide on whether AI-generated text counts as plagiarism.
Frequently asked questions
What are the seven types of plagiarism?
The commonly cited list is direct (verbatim) plagiarism, mosaic/patchwork plagiarism, paraphrasing plagiarism, self-plagiarism, source-based plagiarism, accidental plagiarism, and complete plagiarism. This grouping is a practitioner framework rather than an official standards-body taxonomy; ORI’s own operational split is simpler — plagiarism of text versus plagiarism of ideas.
Is self-plagiarism actually plagiarism?
It depends on which framework is asking. Under 42 CFR Part 93, the U.S. federal research-misconduct regulation, self-plagiarism does not meet the definition of plagiarism, because that definition is scoped to appropriating another person’s work. Under most journal and publisher ethics policies, including COPE’s guidance on text recycling, undisclosed reuse of one’s own prior text or data is treated as a publication-ethics violation regardless of the federal misconduct question. See Self-plagiarism for the full distinction.
Does accidental plagiarism still count as misconduct?
Standard definitions of plagiarism, including the federal one, turn on whether credit was given, not on whether the author intended to deceive. Accidental plagiarism can still meet the definition of plagiarism; institutions vary in how heavily they weigh demonstrated lack of intent when deciding on consequences, but a lack of intent does not, by itself, mean a passage was not plagiarized.
Can plagiarism-detection software catch every type?
No. Similarity-checking software works by string-matching against a reference corpus, so it is effective against verbatim, mosaic, and complete plagiarism, and against some source-based cases, but it cannot detect thoroughly paraphrased, uncredited ideas — there is no matching string to find. It also cannot, on its own, catch self-plagiarism of previously unpublished work, or data/image plagiarism outside its text corpus.
Is reusing my own images or data self-plagiarism?
Presenting one’s own previously published data, figures, or images as newly generated, without disclosure, falls under the same self-plagiarism/text-recycling concern as reused prose — the question editors and institutions ask is whether the reuse is disclosed and whether it misrepresents the work’s novelty, not whether the reused material happens to be text.
Related CASRAI resources
- Plagiarism — the core operational definition
- Textual Plagiarism — verbatim, mosaic, and patchwriting in depth
- Self-plagiarism — legal status vs. publication-ethics status
- Anti-Plagiarism Software — how Turnitin, iThenticate, and Crossref Similarity Check work
- Research Integrity Fundamentals — fabrication, falsification, and plagiarism together
- Research Integrity & Compliance — the full cluster hub
