A reference list is the first place a reviewer looks for evidence that a manuscript was prepared carefully, and the first place they find evidence that it was not. It is also the part of a paper most likely to have been assembled across several years, several reference managers, and several co-authors with different habits.
Three styles cover most of what crosses our desk. This is what actually separates them, in the order the differences will bite you.
The one structural difference
Vancouver and AMA are numeric: sources are numbered in the order they first appear, cited by that number in the text, and listed in that same numeric order at the end. APA is author–date: sources are cited by surname and year, and listed alphabetically.
Everything else in this article is detail. This is the difference that determines how much work a conversion is. Numeric to numeric — Vancouver to AMA, say — is largely a formatting pass. Numeric to author–date, in either direction, restructures the reference list, changes every in-text citation, and can change the order of sentences where an author wrote around a citation number.
In the text
Vancouver uses Arabic numerals, in parentheses or square brackets or superscript depending on the journal — the ICMJE recommendations leave the presentation to the publisher, which is why two Vancouver journals can look different. Numbers are assigned in order of first appearance and reused thereafter.
AMA is more prescriptive: superscript Arabic numerals, placed outside commas and full stops but inside colons and semicolons. Multiple citations are comma-separated with no spaces, and three or more consecutive numbers are condensed with a hyphen: 2-4 rather than 2,3,4.
APA uses author and year: (Okafor, 2021) or Okafor (2021) found…. Two authors are always both named, joined by an ampersand inside parentheses and by and in running text. Three or more authors are cited as Okafor et al., 2021 from the very first citation — a change from the sixth edition, and still one of the most common errors we see in manuscripts prepared from older templates.
How many authors before et al.
This is the single most common conversion error, because the cutoff differs in every direction and applies to the reference list rather than the citation.
- Vancouver: list the first six authors, then et al.
- AMA: list up to six; with seven or more, list the first three, then et al.
- APA: list up to twenty authors in full. With twenty-one or more, list the first nineteen, an ellipsis, then the final author.
Read those again in the direction you are converting. A paper moving from APA to AMA has reference entries with fifteen names that need to become three and et al. A paper moving the other way cannot recover the names that were dropped — they have to be looked up again, one entry at a time, which is why an APA conversion late in a submission cycle is genuinely expensive.
Author names and initials
The numeric styles use initials without full stops and without spaces: Okafor DA. APA punctuates and spaces them, and inverts every name: Okafor, D. A. APA also uses an ampersand before the final author in the reference list; Vancouver and AMA use neither an ampersand nor the word and, just commas.
Titles and capitalisation
Article titles take sentence case in all three — only the first word, proper nouns, and (in the numeric styles) the first word after a colon are capitalised. This trips up authors whose reference manager imported titles in title case from a publisher's metadata, which is most of them.
Journal titles are where they part company. Vancouver and AMA both abbreviate journal titles according to the NLM catalogue — N Engl J Med, J Clin Oncol — with AMA italicising the abbreviation and Vancouver typically not. APA does the opposite in both respects: the journal name is given in full, in italics, in title case, along with the volume number in italics.
Abbreviation is not something to do from memory. Two journals with similar names abbreviate differently, and a wrong abbreviation is indistinguishable from a wrong citation to anyone checking the reference.
Volume, issue, pages
The numeric styles compress: 2021;384(9):823-831, with the year, semicolon, volume, parenthesised issue, colon, and page range. AMA and Vancouver both commonly elide the second page number where the digits repeat — 823-31 — though journals vary and many now ask for full ranges.
APA spaces everything out and drops the labels: 384(9), 823–831 — volume italic, issue in parentheses not italic, full page range with an en dash rather than a hyphen. That last detail matters more than it looks: numeric styles use a plain hyphen in page ranges, APA uses an en dash, and a global find-and-replace in either direction will damage hyphenated surnames if it is not scoped to the page field.
DOIs and URLs
Current practice has converged more than the rest of these styles have. APA 7 presents the DOI as a full URL — https://doi.org/10.1000/xyz — with no doi: label and no full stop after it. AMA 11 also favours the full URL form. Older Vancouver examples still show the bare doi:10.1000/xyz form, and journals differ on which they want.
What matters more than the presentation is whether the DOI resolves. A reference list where every entry is beautifully formatted and one DOI in nine points at the wrong paper is worse than one that is untidy and correct — untidy gets copyedited, wrong gets published. This is the part of reference checking that cannot be done by eye at all, and the reason we verify every entry against Crossref and PubMed rather than reading them.
The traps in an actual conversion
Assume the reference list is not the only thing that changes.
- In-text renumbering. Adding one citation to a numeric manuscript renumbers everything after it. If the citations were typed as plain text rather than inserted as fields, every subsequent number is now wrong, silently.
- Sentences written around citation numbers. As shown in (12) and (13)… becomes ungrammatical in APA. The sentence has to be rewritten, not reformatted.
- Orphans and duplicates. A reference in the list that nothing cites, or a citation pointing at a number that does not exist. Both are invisible in a converted document and both are caught by reconciling the two directions.
- The same source entered twice with different metadata — one from a PDF import, one typed by hand — appearing as two numbers for one paper.
- Reference manager field codes that survive the conversion and re-render the old style when the file is opened on another machine. Unlinking the citations before conversion is the fix, and it must be done on a copy.
- Non-journal sources. Preprints, datasets, software, standards, theses, and clinical trial registrations have specific patterns in each style, and reference managers handle them badly. These are worth checking by hand in any list.
- Secondary citations. A source cited inside another source has a distinct form in each style, and is often better resolved by finding and citing the original.
A practical order of operations
If you are doing this yourself, this sequence produces the fewest reworks:
- Fix the underlying records in the reference manager first, not the formatted output. Every correction you make in the manuscript will be overwritten the next time the bibliography refreshes.
- Deduplicate before converting, not after.
- Verify DOIs while the records are still structured data — it is far easier than reading a formatted list.
- Convert, then reconcile in-text citations against the list in both directions.
- Only then run the presentation pass: names, abbreviations, capitalisation, dashes, spacing.
- Read the first and last five entries by hand. Errors cluster at the ends, where the list was started and where it was rushed.
Where this ends up
Most of this is rule-following, and rule-following is what programs are for. We keep the target journal's requirements as a style sheet, rebuild the list against it entry by entry, check every identifier against Crossref and PubMed, and reconcile the citations with the list in both directions. What comes back is tracked, with a log of every entry that changed and every entry we could not verify.
The judgment calls are the ones left over: whether a preprint should be cited at all now that the version of record exists, whether a secondary citation should be replaced with the original, whether a reference is doing the work the sentence claims. Those go into the query memo, for the author to answer.
Related service