Learn better
How to read a research paper without losing the argument
Start with a question you need the paper to answer. Scan its purpose and structure, then follow one important claim from the results back to the method. Finish with a short note that separates the finding, the authors’ interpretation and what remains uncertain. You do not have to understand every detail on the first reading.
Key takeaways
- Read for a specific task, rather than highlighting everything.
- A reported result and a causal explanation are different claims.
- Keep a page, table or section location beside each useful finding.
- Return to the original paper before citing a summary.
Decide what you need from this paper
I would begin by writing one sentence: “I need this paper to help me explain …”. A seminar discussion, an assignment claim and an attempt to reproduce a method require different depth. Set a stopping point, such as being able to explain one result and one limitation, rather than promising to finish in a fixed number of minutes.
First identify the article type. An original study reports its own investigation; a review brings together other work; a commentary argues a position. Their useful sections differ. The workflow below concentrates on an empirical paper with methods and results. For a theoretical paper, trace the assumptions and reasoning instead of looking for participants.
Check that you have the complete paper, not just its abstract. Record its title, authors, publication year and DOI or stable link. Use your library’s access or a legitimate open version when available. Check the publisher’s page for a correction or retraction notice before relying on it.
Read in passes, with a job for each one
S. Keshav’s three-pass approach moves from an overview to deeper understanding. I use that broad principle here, with a narrower student task: decide whether a paper is relevant, inspect a claim you might use, then verify the parts your assignment depends on. This is a practical workflow, not a tested promise of faster reading or better grades.
- An unfamiliar term that changes the meaning needs checking now; a minor detail can go on a questions list.
- If the topic is unfamiliar, read an introductory source and return.
- If you cannot explain the method behind a claim, keep that uncertainty in your notes.
| Pass | What to inspect | Useful output |
|---|---|---|
| Find your route | Title, abstract, headings and conclusion; then locate the method and main figures. | One sentence on the question and whether it fits your task. |
| Follow a claim | One result, its figure or table, and the method that produced it. | A finding, its source location and a limitation. |
| Check before using | The relevant details, definitions, analysis and competing explanations. | A defensible summary, or a precise question to resolve before citing. |
Follow the evidence, not just the abstract
Carey, Steiner and Petri recommend identifying your reading goal, understanding the article type and actively examining the figures and tables. Their guidance is expert advice about reading, not an experiment proving one reading order is best.
For a finding you might cite, ask what was measured, in whom or what, under which conditions, and against what comparison. Read the figure caption and relevant method together. A chart of confidence ratings does not measure examination performance; a group difference does not by itself explain why the groups differ.
Treat words such as “improves”, “causes” and “proves” as prompts to inspect the design. Ask whether another explanation fits the result. Separate a result reported by the paper from your own proposed application. A limitation does not automatically make a paper useless; it defines what you can reasonably use it to support.
Worked example: does a note app improve learning?
The following mini-study is invented for this guide. Its students, app and results are not a real experiment or evidence for a product recommendation.
Suppose 40 volunteers choose between a note app and paper notes while studying the same chapter. Twenty choose each method. The app group reports an average confidence of 4.2 out of 5; the paper group reports 3.5. The app group also spends 30 minutes studying, compared with 20 minutes for the paper group. There is no pre-test and no later knowledge test.
A tempting summary is: “The app improved learning by 20%.” That is not supported. The 0.7-point difference is in reported confidence. Dividing 0.7 by 3.5 gives 20%, but that arithmetic does not turn a rating into a measure of learning. The groups chose their methods and spent different amounts of time studying.
| Question | What the example tells you | What it does not establish |
|---|---|---|
| What changed? | Average reported confidence differed by 0.7 points. | A difference in retained knowledge. |
| Were conditions matched? | The chapter was the same. | Equal study time, starting knowledge or random allocation. |
| What can I write? | App users reported higher confidence in this small self-selected sample. | That the app caused better learning or will help every student. |
| What would I want next? | A clear knowledge measure, comparable conditions and a stronger design. | A guaranteed result from repeating the same observation. |
Turn the reading into a traceable note
My note for that fictional example would be: “In a self-selected group of 40 volunteers, app users reported higher confidence after reading. The groups studied for different lengths of time, and no knowledge test was used. The example cannot establish an effect on learning.”
For a real paper, add the exact table, figure, page or section where each part comes from. Keep direct quotations visibly quoted and separate from your paraphrase. Reopen the source before using the note in an assignment. A saved DOI helps you find the paper; it does not identify the particular result you used.
If you are working through a whole chapter rather than a single study, use the textbook note workflow to organise claims and source locations.
A worksheet for your next paper
Copy these prompts into a notebook or document, or print this page. The worksheet is an original aid for organising your reading. It is not a scoring system for research quality.
- My reading question and assignment purpose:
- Paper title, authors, year, DOI or stable link:
- Article type and current publication status:
- Question investigated and why it matters:
- Method, sample or material, and comparison:
- One result I may use, with its figure/table/page:
- What was measured, and what was only inferred:
- Important limitation or alternative explanation:
- My summary in two or three sentences:
- Unresolved term, method or question, and next action:
What if the paper still makes little sense?
Write the smallest question you can identify: “What does this axis measure?” is more useful than “I do not understand the paper.” Check its definitions, caption and relevant background source. If a statistical method or specialist concept remains central and unclear, ask a tutor or librarian for the right starting resource. Do not hide that gap by copying the conclusion.
An AI summary can suggest questions or explain terminology, but I would not cite its account of a finding without checking the paper. Verify numbers, quotations and references in the original. Follow your institution’s rules and the material’s access conditions before uploading it to a third-party service.
Finish with a claim you can explain
Close the paper and explain the research question, one finding and one limitation in your own words. Then reopen it and correct your explanation. Being able to repeat the abstract is less useful than knowing which claim the evidence supports.
Use the student editing checklist when you turn that note into an assignment paragraph. Check that your sentence stays within the study’s population, measure and design.
For keeping useful ideas over time, continue with remembering what you read. Reading, checking and later recall serve different parts of the task.
Sources and further reading
- S. Keshav: How to Read a Paper (three-pass reading guidance) (accessed 28 September 2026)
- Carey, Steiner and Petri (2020): Ten simple rules for reading a scientific paper (accessed 28 September 2026)
Access dates are recorded beside each source. Please report a source or factual concern.