Research papers increasingly depend on data produced by others: survey results, genomic sequences, climate records, and clinical datasets that other teams collected, cleaned, and shared. When your argument rests on such data, the dataset deserves a proper citation, just as a journal article does. A citation gives credit to the data producers, lets readers find the exact data you used, and makes your own work reproducible.
This guide explains how to cite datasets correctly. It covers what counts as a citable dataset, how to find its identifier and citation details, how to format the citation in APA, MLA, and Chicago styles, and how dataset citations relate to data availability statements and to citations of software and code. Follow it, and your data citations will be as careful as your article citations.
Data production is research work. Collecting, cleaning, documenting, and sharing a dataset can take years, and the researchers who do it deserve credit that is visible and countable. A formal citation in your reference list gives them that credit in a way that a passing mention in the acknowledgements does not.
Citation also serves readers. A reader who wants to check your analysis or reuse the data needs the exact dataset you used, not a similar one or a homepage that has since changed. A complete citation with a persistent identifier points them to the right version and protects your paper against the charge that its evidence cannot be traced.
A citable dataset is a defined body of data that has been deposited in a repository or archive, given a stable identifier, and documented well enough for others to understand and reuse. Examples include survey data deposited in a data archive, sequences deposited in a nucleotide database, and observational records published through a domain repository. The key test is whether a reader can locate and identify the exact data from your citation alone.
Not everything called data is a dataset in this sense. A table copied from a website, a figure from a report, or data mentioned in passing in another paper should usually be cited through the publication that presents them, not as a standalone dataset. If the data have their own repository record and identifier, cite the dataset; if they exist only inside a publication, cite that publication and say where the data appear in it.
Start at the repository record, not at a search engine result. A dataset deposited in a reputable repository — such as Zenodo, Dryad, Figshare, or a domain archive — has a landing page stating the title, creators, publication year, version, publisher or repository, and a persistent identifier, usually a DOI. Copy these details exactly as the repository states them, including the version number.
Many repositories also suggest a citation format on the landing page. Use it as your starting point, then adapt it to your journal's style. If the repository gives no suggested citation, build one from the stated details: creators, year, title, version, repository, and identifier. If you cannot find an identifier or a stable landing page, say so to your editor, and consider whether the data are documented well enough to support a citable claim at all.
In APA style, a dataset reference follows the same logic as other references: author, date, title, and source, with the version and identifier added. A typical entry names the creators in author position, gives the year of deposit in parentheses, italicises the dataset title, adds the version in brackets where relevant, names the repository as publisher, and ends with the DOI as a link. Check the current APA guidance for the exact punctuation, since style manuals are revised.
In the text, cite the dataset by creator and year, as you would an article, distinguishing same-year datasets with letters as APA does for articles. In your methods, mention the dataset by name and version too, so a reader who skips the reference list can still identify what you used. Our guide on citing AI tools in research papers covers a related citation problem with its own style rules.
MLA style treats a dataset as a work with creators, a title, a container, and a location. A typical entry begins with the creators, gives the dataset title in italics, names the repository as the container, adds the version and date where available, and ends with the DOI or stable URL. The order and punctuation follow MLA's core-elements logic, so a dataset entry looks like other MLA entries, with the repository playing the role of publisher.
In the text, MLA uses parenthetical citations keyed to the works-cited entry, usually by creator name. If the dataset has many creators, follow MLA's rule for the number of names shown in the entry and in the citation. As with other styles, the goal is unambiguous identification: a reader should be able to move from your in-text citation to the exact dataset record without guessing.
Chicago style offers two systems, and you should use the one your journal specifies. In the notes-and-bibliography system, a dataset appears in a footnote at first mention and in the bibliography, with creators, title, version, repository, date, and identifier. In the author-date system, the reference list entry resembles the APA pattern, and in-text citations give creator and year.
Whichever system you use, include the version and the identifier, since datasets are frequently updated and a citation without a version may point to different data than the ones you analysed. If the repository assigns a new DOI to each version, cite the DOI of the version you used, not the general DOI of the dataset series. State the version in your methods as well, so the analysis and the citation agree.
Treat dataset citations with the same discipline as article citations. Every dataset mentioned in the text should appear in the reference list, and every dataset in the reference list should be cited in the text. Do not list datasets you considered but did not use, and do not mention a dataset in the text without a corresponding reference entry.
Be consistent about how you name datasets in the text. Use the dataset's title or a clear short form, and use it the same way throughout. If you analyse several versions or subsets, name each one distinctly and say how they differ. A paper that refers vaguely to 'the data' while its reference list contains several datasets forces readers to guess which data support which claim, and that guessing undermines confidence in the whole paper.
A data availability statement and a dataset citation do different jobs, and most journals now want both. The statement tells readers where the data behind your paper can be found, including your own newly generated data. The citation gives formal credit for data produced by others that you reused. If you reused an existing dataset, cite it in the reference list and mention it in the statement as well, so the two point to the same record.
For your own data, deposit them in a suitable repository before submission where your field expects it. A statement that says data are available on request is weaker than a repository link, and some journals no longer accept it. Our guide on writing a data availability statement explains the wording for each situation, including restricted and third-party data.
Datasets are often analysed with software and code that also deserve citation. Cite the software package by its creators, version, and identifier where one exists, and cite code deposited in a repository in the same way as a dataset. If your analysis code is central to the results, deposit it, cite it, and link it to the dataset citation, so readers can follow the full chain from data to result.
Keep software, code, and data citations distinct: a citation to a software manual is not a citation to the data it processed, and vice versa. List each in the reference list under its own entry, and mention each in the methods where it was used.
When you adapt or derive new data from an existing dataset — by cleaning, combining, or reanalysing it — cite the original dataset and describe what you changed. Readers need to know which parts of your data are yours and which parts came from elsewhere, and the original producers deserve credit for the underlying work even when your analysis is new. State the derivation in your methods, not only in a footnote.
If you combine several datasets, cite each one separately rather than merging them into a single vague reference. Check the licence on the repository record before adapting data, since some licences require attribution in a specific form or prohibit certain uses, and record the terms with your analysis notes.
Before submitting, verify every dataset citation against its repository record: creators, year, title, version, repository, and identifier. Click each DOI or link and confirm it resolves to the dataset you used. Check that in-text citations match the reference list and that version numbers agree across text, statement, and reference entry.
Ask a colleague to trace one dataset citation from your paper to its repository record without your help. If they cannot, the citation is not yet good enough. This check takes minutes and prevents a class of reviewer complaint that is entirely avoidable. Journals increasingly check data citations during screening, and a paper with broken or vague dataset references may be returned before review, however strong its analysis.
Styles change between editions, so confirm the current rules of the style your journal requires. The table below gives the usual pattern.
| Style | In-text citation | Reference entry pattern | Version and identifier |
|---|---|---|---|
| APA | Creator and year | Creators (year). Title [dataset]. Repository. DOI | Version in brackets; DOI as link |
| MLA | Creator in parentheses | Creators. Title. Repository, version, date. DOI | DOI or stable URL at the end |
| Chicago notes | Footnote at first mention | Creators, Title, version, repository, date, DOI | Version stated; DOI given |
| Chicago author-date | Creator and year | Creators (year). Title. Repository. DOI | Version stated; DOI of the version used |
Dataset citations fail most often in a few predictable ways: citing a repository homepage instead of the dataset record; omitting the version, so the citation points to different data than those analysed; listing a dataset in the references without citing it in the text; and citing data through a paper that merely mentions them, instead of citing the dataset itself. Each of these breaks the chain a reader needs to verify your work.
Another common fault is confusing a data availability statement with a citation. The statement says where data can be found; the citation gives credit and identification in the reference list. You usually need both. Finally, never cite a dataset you have not examined: a citation asserts that you used those data, and a reviewer who checks will discover quickly whether that assertion is true.
Run through this list for every dataset in your paper before submission:
These authoritative external resources explain the standards and tools mentioned in this guide:
Should I cite a dataset with a DOI or a URL?
Use the DOI where one exists, because a DOI is a persistent identifier designed to keep resolving even if the repository's web address changes. Use a stable URL only when no DOI exists, and make sure it points to the dataset record, not to a repository homepage.
How do I cite my own dataset?
Cite it like any other dataset, with yourself as creator, the deposit year, title, version, repository, and DOI. Citing your own data is good practice: it makes the data findable and gives your data-sharing work visible credit. Mention it in your data availability statement as well.
What if a dataset has no DOI?
Cite it with the most stable identifier available, such as a repository accession number or a stable URL, together with the full creator, title, version, and repository details. If no stable record exists at all, discuss the situation with your editor, since the data may not be documented well enough to support a citable claim.
Do I need to state the dataset version?
Yes, whenever the dataset is versioned. Datasets change, and a citation without a version may point to different data than the ones you analysed. Give the version in the reference entry and in your methods, and cite the DOI of the exact version you used.
How do I cite a large database rather than a single dataset?
Cite the database as a whole if you used it as a resource, and cite the specific dataset or query result if you analysed a defined subset. Say in your methods what you extracted, when, and under what criteria, so readers can reconstruct the data you actually used.
Is a data availability statement the same as a dataset citation?
No. The statement tells readers where the data behind your paper can be found; the citation gives formal, reference-list credit for data you reused. Most journals now expect both, and the two should point to the same records.