How to Read a Research Paper Efficiently
Nobody who reads papers for a living reads them the way you were taught to read a book. They triage in ten minutes, and only a few papers a month ever earn a full read. Here's the method, the reading order the evidence actually supports, and the note-taking habit that keeps a citation usable months later.
The mistake: reading front to back
Opening a PDF at the title and grinding through to the references is slow, and it front-loads the two sections you are least equipped to judge. It is also not what researchers do. In a 2024 PLOS ONE survey by Shiely and colleagues at University College Cork, science and health researchers reported that they do not read IMRAD papers in printed order — 98.6% read the abstract first, and what they read next was predicted by how important and how easy they judged each section to be.
The gap between novices and experts is not speed. It is knowing which paper deserves an hour, and which one deserves eight minutes and a note.
The three-pass method
The most widely used framework for this comes from S. Keshav's 2007 note How to Read a Paper in ACM SIGCOMM Computer Communication Review. The idea: read a paper in up to three passes, each with its own goal and time budget, and stop as soon as the paper has told you what you needed.
Pass 1 — triage (5–10 minutes)
Keshav's first pass is five steps:
- Read the title, abstract and introduction carefully.
- Read the section and sub-section headings — and nothing else.
- Glance at the mathematical content to see what it rests on.
- Read the conclusions.
- Skim the references, ticking off the ones you already know.
At the end of it you should be able to answer the five Cs: Category (what kind of paper is this — measurement, prototype, analysis?), Context (what other work is it in conversation with?), Correctness (do the assumptions look sound?), Contributions (what does it actually claim to add?) and Clarity (is it well written?).
Then make the decision the pass exists for: stop, file it, or read on. Most papers should stop here — that is the method working, not you being lazy.
Pass 2 — comprehension (up to an hour)
Now read the body, ignoring proofs and fine detail. Two things earn most of your attention here:
- The figures and tables. Keshav is specific: check that axes are properly labelled, and that results are shown with error bars so you can see whether a difference is statistically meaningful. A surprising number of confident claims fall apart at this step.
- The references you haven't read. Mark them as you go. This is how one paper turns into a reading list.
After a second pass you should be able to summarise the paper — claim, evidence, main limitation — to someone else. If you can't, the paper is either badly written or outside your background; both are useful things to know.
Pass 3 — depth (about an hour, or a whole afternoon)
The third pass is for papers you will build on, replicate, or argue with. Keshav describes it as attempting to virtually re-implement the paper: make the same assumptions the authors made and recreate the work in your head, which surfaces the hidden assumptions their own write-up glosses over. He budgets about an hour for an experienced reader and four or five hours for a beginner. You will do this for a handful of papers in a project, not for fifty.
Read the results before you trust the claim
Hubbard and Dunbar surveyed 260 undergraduates and researchers in Biological Sciences at a research-intensive UK university (PLOS ONE, 2017) and found that what people value in a paper shifts with career stage: undergraduates ranked the abstract as the most important section, while postgraduate researchers ranked the figures and tables highest. Inexperienced readers also found the methods and results hardest to read, and skill at reading results developed slowly across a career.
The practical lesson is blunt: the abstract is the authors' interpretation of their data, and it is the section most likely to overstate. If a finding is going to carry weight in your own argument, look at the figure and the sample size before you write the sentence that cites it.
A reading order that works
- Abstract — is this even about my question?
- Figures and tables — what did they actually find?
- Discussion and conclusion — what do they say it means, and what do they admit it doesn't?
- Introduction — only if the framing matters to you; this is also where you harvest citations.
- Methods — when the claim matters, or when the result looks too clean.
Take the note now, or you'll reread the paper
The expensive mistake is not slow reading — it is reading a paper properly in March and having to open it again in June because your note said "good on motivation, check p. 4". For anything you might cite, capture four things while the paper is still open: the claim in your own words, the evidence behind it (n, design, effect), the page or section, and your own one-line verdict on how you'll use it.
Do the reference at the same time. A citation captured at the moment of reading is accurate; one reconstructed from memory a month later is how wrong page numbers and half-remembered author lists get into bibliographies.
Paste the details into our free citation generator and get a correctly formatted APA, MLA, Chicago, Harvard or IEEE entry to drop straight into your notes — before you close the tab.
Six questions to ask every paper
- What exactly is the claim — and how narrow is it really?
- What would have to be true for this to be wrong?
- Who was studied, how many, and for how long?
- Is the comparison fair, or is the control doing the work?
- What do the authors themselves list as limitations?
- Does anything here change what I was going to write?
That last one is the point of reading at all. A paper you can't connect to your own argument is a paper you have filed, not read.
Turning reading into a literature review
Keshav's method scales up neatly. Start with a few recent, relevant papers and give each one a first pass. Watch for the references and author names that keep reappearing across them — those are your key papers, and they get a proper second pass. Then look at where those authors publish and scan the recent proceedings or issues of those venues. When new papers stop producing new names, you have found the edges of the conversation. Our guides to searching Google Scholar and building an annotated bibliography pick up from there.
Write from papers you've actually read
CiteGood drafts from real sources and formats every citation for you — so what ends up in your bibliography is something you can open, check and defend.
Try CiteGood free →Frequently asked questions
How long should it take to read a research paper?
It depends on which pass you are doing. Keshav's (2007) three-pass method budgets five to ten minutes for a first-pass triage, up to an hour for a second pass that gets you the argument and the evidence, and about an hour for a third, full-depth pass by an experienced reader — four or five hours if the area is new to you. Most papers should never get past the first pass.
In what order should you read the sections of a paper?
Not the order they are printed in. A 2024 PLOS ONE survey by Shiely and colleagues found that science and health researchers do not read IMRAD papers linearly: 98.6% read the abstract first, and after that the order is driven by how important and how difficult they judge each section to be. A practical order is abstract, then figures and tables, then discussion, then introduction, then methods.
Should I read the methods section at all?
Read it whenever the paper's claim matters to your argument — the methods are where you find out whether the result means what the abstract says it means. Hubbard and Dunbar (2017) found that inexperienced readers rate methods and results as the hardest sections and undervalue the results, which is exactly why skipping them is a habit worth breaking early.
How many papers do I need to read for a literature review?
There is no fixed number, and counting is the wrong target. Work outward from a handful of recent, relevant papers: first-pass each one, note the references that keep reappearing across them, and read those key papers properly. You stop when new papers stop producing new names — not when you hit a quota.
Can I use AI to summarise papers I haven't read?
Use it to triage, not to substitute. An AI summary is a reasonable way to decide whether a paper is worth your first pass, but you cannot cite a claim you have only met in a summary — the summary may compress away the caveat that makes the claim false for your case, and models also invent references outright. Open the paper before the claim goes in your draft.