How to read a research paper (and what to skip)
Nobody reads research papers the way reading guides describe. What actually happens is a three-minute triage, a look at the figures, and a decision. The useful question is not how to read a paper but which of three jobs you are doing, because screening a paper for relevance, extracting what it found, and deciding whether to believe it cost three minutes, twenty minutes and forty minutes respectively, and doing the forty-minute version on a paper that deserved the three-minute one is how a reading list becomes impossible. This page gives you the budgets, what to skip and what skipping it costs, and the routine for a paper you only need one fact from.
- Decide which of three jobs you are doing
- The three-minute screen
- What to skip, and what skipping it costs
- Reading a paper you need one fact from
- The trust pass, and when to spend it
- Where the three-pass method fails a student
- Notes that survive to the draft
- Getting through a list of thirty
- Keeping the read-and-cite loop honest
To make the advice below concrete rather than impressionistic, we measured the thing being read. On 11 August 2026 we took the full text of 80 open-access research articles from PLOS ONE, indexed under Climate change and Transportation and published since January 2024, and counted the words in every section. The median article ran 5,912 words of body text with a 238-word abstract, 5 figures, 3 tables and 52 references. At the 198 words a minute this blog uses to estimate its own read times, straight-through reading of that median paper is about half an hour, before you have understood anything, and technical prose is slower than blog prose. That arithmetic is why selective reading is not laziness. It is the only way the numbers work.
Decide which of three jobs you are doing
Before you open a paper, answer one question: what do I want from this? There are only three honest answers, and each has a different budget.
- Screening. Is this paper relevant to my question at all? Three minutes. Most papers on a reading list never get past this, and that is the correct outcome.
- Extracting. What did this paper find, well enough that I can describe and cite it accurately? Fifteen to twenty minutes.
- Trusting. Should I build an argument on this finding? Thirty to forty-five minutes, because this is the one where you have to read the methods.
The mistake that eats a week is doing job three on papers that needed job one. The second mistake, which costs marks rather than time, is doing job one and writing as if you had done job two, citing a paper for a finding you took from its abstract.
The three-minute screen
Read these four things, in this order, and then decide. Do not scroll to the top and start.
- The title and abstract (60 seconds). The median abstract in our sample was 238 words, under 4 percent of the paper, and it contains the question, the approach and the headline result. If the abstract does not touch your question, you are done.
- The figure captions (60 seconds). Not the figures, the captions. In an empirical paper the findings live in the figures, and the caption states what each one shows. Five captions tell you what the paper actually did faster than any paragraph.
- The last paragraph of the discussion (30 seconds). This is where authors say what they think it all means, in plainer language than anywhere else in the paper, and usually with the hedges they left out of the abstract.
- The limitations (30 seconds). Usually the paragraph before the conclusion. It is the highest-value paragraph per word in the whole paper, because it tells you exactly what you can and cannot cite this paper for.
Then make one of three decisions: discard it, file it as background you will cite once, or promote it to a real read. Write the decision down next to the reference. A paper you screened and did not record is a paper you will screen again in a fortnight.
What to skip, and what skipping it costs
Skipping works when you know the price. Here is the price, in the words we counted.
| Section | Share of the paper | Skip it when | What you lose |
|---|---|---|---|
| Methods | 26% | You are not judging or repeating the study | The ability to say whether the finding is any good |
| Introduction | 14% | You already know the field | Mostly a longer restatement of the abstract, plus the citations it hangs on |
| Results prose | 21% | The figures and captions carry it | The exact statistics, which you need only if you quote a number |
| Discussion | 23% | Never skip entirely | The authors' own account of what the result means and does not mean |
| Conclusion | 4% | You have read the discussion | Almost nothing. A median of 190 words, usually a summary |
Methods is the biggest single saving available, at roughly a quarter of the paper, and it was longer than the results section in 33 of the 58 papers that carried both. That surprises most students, and it changes how you triage: the dense middle you have been grinding through is not where the findings are, it is where the justification for them is. Read it when you need the justification.
The related work or literature review section, which 18 of our 80 papers carried, is worth almost nothing to read and a great deal to mine. Scan it for the papers it cites approvingly, note two or three, and move on. It is a map of the field written by someone who read it for you.
Reading a paper you need one fact from
This is the most common reading job in student work and no guide covers it. You have written a sentence, or you are about to, and you need the number or the finding that supports it. You do not need to understand the paper. You need to be sure of one claim in it.
- Search, do not read. Open the PDF and search for the unit, the number, or the distinctive term. Search the tables too: the median paper in our sample had three of them, and a specific value is more often in a table than in prose.
- Read the sentence before the one you found. The qualification that makes a number honest is almost always in the preceding sentence: the sample it applies to, the conditions, the model it came out of. Take that sentence with you or you will cite a general claim the paper never made.
- Check whether the paper owns the fact. If the sentence carries a citation of its own, the paper is repeating someone else's result. Follow it and cite the original. Citing a paper for something it cited is how errors propagate, and it is trivially caught by anyone who checks.
- Compare the abstract and the discussion. If the claim is stronger in the abstract than in the discussion, cite the discussion's version. Abstracts overstate; discussions hedge; the hedged version is the one the evidence supports.
- Record it now. Author, year, DOI, the page or figure number, and the exact sentence in quotation marks. Ten seconds here saves the twenty minutes of re-finding it that everyone budgets zero for.
Done properly this takes four minutes and produces a citation you can defend. Done by memory a week later, it produces the single most common citation failure there is: a real paper attached to a claim it does not quite make.
The trust pass, and when to spend it
Spend forty minutes on a paper when your argument leans on it, when it is the only source you have for a claim, or when the result is surprising. Then read for the join between the claim and the evidence, which is the only thing the extra time buys.
Four questions do most of the work. What exactly was measured, and is it the thing the conclusion talks about? How many, of what, and would the result plausibly hold outside that group? Does the discussion promote a correlation into a cause somewhere between the results section and the abstract? And what do the authors themselves say the study cannot show?
That last one is not a formality. A paper that names its own limits is telling you the precise claim it will support, which is also the claim you can safely make with it behind you. Judging the source itself, the venue, the peer review, whether the journal is real, is a different skill and we cover it in how to find credible sources.
Where the three-pass method fails a student
Almost every "how to read a paper" article on the internet descends from one two-page note: S. Keshav's How to Read a Paper, published in ACM SIGCOMM Computer Communication Review 37(3), pages 83 to 84, in July 2007. It is a genuinely good piece of writing and worth the ten minutes it takes to read in full. It is also aimed at a reader who is not you.
The note's own numbers say so. The first pass is "about five to ten minutes". The second "should take up to an hour". The third "can take about four or five hours for beginners, and about an hour for an experienced reader", and its stated purpose is "to attempt to virtually re-implement the paper", which Keshav recommends "particularly if you are reviewer". The intended reader is a beginning graduate student who reviews conference submissions and is building a research programme.
Run those budgets against a real assignment. Twelve sources, second pass on each, is twelve hours of reading before a word is written. Three third passes is another fifteen. That is not a coursework schedule, and treating it as one is why students who follow the advice faithfully end up further behind than students who ignore it.
What the note gets right, and what almost every derivative version drops, is the escalation principle: read in passes and stop as soon as you have what you need, spending effort only where it pays. It also says the thing most reading advice will not, that a first pass is a complete and legitimate reading for a paper outside your area, and that most readers of any paper make one pass over it and no more.
What it does not have is the two passes a student needs most. There is no three-minute screen for a reading list of thirty, and there is nothing at all about the one-fact case above, because a researcher building a field does not have that problem and someone writing an essay on Thursday has almost nothing else.
Our own view, since these pages are usually written as if nobody has one: the three-pass method is the right idea with the wrong budget for coursework, and if we could delete one reading habit it would be starting at the introduction. It is 14 percent of the paper, it largely restates the abstract at four times the length, and it is the section that makes students feel they have begun reading when they have not yet learned anything the abstract did not tell them. Start at the figures instead.
Notes that survive to the draft
The reading is wasted if the trail back is broken, and this is where most of it is wasted. A note that records a finding without recording where it came from cannot become a citation without re-reading the paper, so it usually becomes a citation from memory instead, which is where misattributions come from.
One rule covers it: no note exists without its source and its location attached. Author, year, DOI, page or figure. Quotation marks around anything you copied verbatim, always, at the moment you copy it, because in three weeks you will not remember which words were yours. A reference manager such as Zotero (free and open source) keeps the paper and the note together and exports a formatted bibliography at the end, but a plain document works if the rule holds. When you write, turning a note into a cited sentence should be transcription, not detective work.
Getting through a list of thirty
Sort before you read. Screen all thirty first, three minutes each, which is ninety minutes and gives you a sorted pile instead of an undifferentiated one. Expect the pile to come out roughly as twenty discarded or filed as background, eight worth a twenty-minute extraction, and two worth the trust pass. That ratio is a working assumption, not a law, but it is the shape most reading lists actually have, because a reading list is assembled to cover a field and your paper answers one question in it.
Read the most recent good review or survey first if the list has one. It is the fastest way to learn which of the other twenty-nine matter, and it will name the papers everyone in the field cites, which are the ones worth your deep read. Keshav's own advice for entering a new field is the same move in reverse: find three to five recent papers, read their related work sections, and look for the citations and author names that repeat.
Finding the papers in the first place is its own job, covered in how to find sources for a research paper, and turning a read stack into an argument is the heart of writing a literature review. Reading is the hinge between them, and it is the one step nobody can do for you, because deciding what a paper means for your argument is the argument.
Keeping the read-and-cite loop honest
The gap between what you read and what you write is where citations go wrong, because memory fills it in and memory is confident. CiteOwl closes that gap by doing the reading before it writes: it searches real academic databases, reads the papers it finds, and ties every claim in your draft to the exact passage that supports it, with the quote shown so you can confirm the source says what the sentence claims. Nothing is cited that was not read. You still do your own reading and own the argument, but the loop stays anchored to passages you can open and check. If you want to test references you already have, our free citation checker resolves them against CrossRef without an account.