How to Read a Scientific Paper: The Order Experts Use

Do not read a scientific paper from start to finish. Read the abstract to decide whether the paper is worth your time, go straight to the figures and results to see what the data actually shows, then read the discussion to see what the authors claim it shows, and consult the methods last.

A research paper is not written in the order it is meant to be read. It is written in the order a journal requires, which is a filing convention rather than a teaching one. Nobody tells students this, so the first assigned paper gets read like a chapter, front to back, and the experience is miserable enough that many people conclude they are not smart enough for primary literature. They are. They are just using the wrong entry point.

Why reading a paper start to finish does not work

Three things go wrong when you start at word one and continue to the end.

You spend your freshest attention on the least useful section. The introduction is a literature review written to justify the study to reviewers. It is the easiest part to read and the least informative, which is a bad combination, because easy reading feels like progress. By the time you reach the part that carries the actual finding, your concentration is gone.

You hit the methods before you know what they are for. Read cold, a methods section is an unmotivated wall of reagents, instruments, and statistical tests. Read after you know what the paper found, the same section becomes a short list of specific questions: how many, compared to what, measured how. Same words, completely different difficulty.

You meet the authors' interpretation before you form your own. This is the expensive one. Once you have read a confident sentence telling you what a result means, it is genuinely hard to look at the figure and see anything else. Order is not just about efficiency here. It decides whether you are evaluating the evidence or being walked through a conclusion.

There is survey evidence that this reordering is what expertise actually looks like. Katharine Hubbard and Sonja Dunbar surveyed 260 undergraduates and researchers in biological sciences at a research intensive UK university and published the results in PLOS ONE as Perceptions of scientific research literature and strategies for reading papers depend on academic career stage. They asked participants to rank the sections of a paper by importance, and found a clean split by seniority: undergraduates ranked the abstract and the discussion as the most important sections, while senior researchers rated those two as relatively unimportant for understanding a paper. Experienced researchers ranked the figures, tables, and results highest, and those were exactly the sections undergraduates undervalued. The methods section was rated least important by undergraduates and rose steadily with career stage.

Read that split carefully, because it is not a difference in reading skill. It is a difference in where people look. Novices weight the two sections where the authors describe their own work. Experts weight the two sections where the evidence sits. The same paper yields a summary to one group and a judgement to the other.

What order should you read a scientific paper in?

Five passes, and you will often stop after two. The point of the order is that each pass gives you the context that makes the next one cheap.

1. Abstract, as triage rather than as a summary

The first pass answers one question: do I read this at all? Give it two minutes and look for the population studied, the intervention or comparison, the headline result, and the size of the study. If the paper does not match what you actually need, stop. Deciding not to read a paper is a skill, not a failure, and it is most of what makes experienced readers look fast.

What the abstract is not is a substitute for the paper. It is the most compressed and most promotional part of the document, written to be indexed and to attract citations. Treat it as a claim you are about to check.

Some researchers go further than that. In a widely circulated guide to reading papers written for non-scientists, the geneticist and anthropologist Jennifer Raff describes using the title and abstract to pick what is worth reading, then reading the abstract last once she sits down with a paper properly. Her reason is the anchoring problem stated plainly: an abstract is a summary of the entire paper, and she does not want to be inadvertently biased by how the authors interpret their own results before she has looked at them. You do not have to adopt that exact discipline to take the point. Use the abstract to choose, not to conclude.

2. Figures, before any body text

Go to figure one and read every figure and caption in order. A well made figure carries the finding more directly than the paragraph describing it, and captions in most journals are written to be self-contained, so the figures plus their captions are close to a standalone version of the paper.

For each figure, name the axes out loud, work out what one data point represents, find the control or comparison condition, and only then look at the difference being highlighted. In the Hubbard and Dunbar survey, figures and tables were ranked more important than the text of the results by every career stage, including the undergraduates who ranked results low overall. Everyone agrees the pictures matter. Novices simply arrive at them last, tired, after the interpretation has already been supplied.

3. Results, checked against the figures

Now read the prose describing those figures, and read it as a cross-check rather than as new information. You are asking one question throughout: does this sentence describe the picture I just looked at? Most of the time it does. When it does not, you have found something genuinely worth noting, and you found it because you looked at the figure first.

4. Discussion, compared against the view you already hold

Read the discussion only after you can state the finding in your own words. Its function is to tell you what the authors think the result means, and it is written persuasively, because the authors are advocating for their own work. That is not misconduct. It is the genre.

Reading it fourth turns it from an instruction into a comparison. Where the discussion says more than you were prepared to say from the figures, you have found the paper's pressure point. Also read the limitations paragraph properly rather than skimming it, since authors frequently disclose the study's real weakness there in one carefully worded sentence.

5. Methods, consulted rather than read

Come to the methods with specific questions produced by the previous passes, and answer only those. If the result rests on a comparison between two groups, find how the groups were assigned. If it rests on a measurement, find how the measurement was made and validated.

This is also the section you skip most often and should skip least often for a paper you intend to rely on. It moved from least important to third most important across career stages in the survey above, which is a fair description of what happens when you start actually depending on results being true.

How do you evaluate the methods if you are not a domain expert?

You can get most of the way with structural questions that require no specialist knowledge at all. A method can be wrong in a domain-specific way you cannot detect, but the more common problems are structural, and structural problems are visible to anyone.

Ask these five, in order:

What was compared with what? Every empirical claim is a comparison. Find the control or comparison condition and confirm it differs from the treatment condition in only the way the paper claims. When there is no comparison group at all, the paper can describe but it cannot attribute.

How many, and of what? Find the sample size and what a single unit actually is. Twelve participants, twelve mice, and twelve measurements on the same three mice are three very different studies, and the third is sometimes reported in a way that reads like the first.

Who ended up in which group, and how? Random assignment supports causal language. Self-selection, convenience sampling, and retrospective grouping do not. This one question separates most overstated findings from defensible ones.

Is the thing measured the thing claimed? Papers routinely measure a proxy and discuss the underlying construct: a questionnaire score standing in for wellbeing, a biomarker for disease progression, an exam mark for learning. The proxy may be reasonable. The substitution should still be visible to you.

Was the analysis decided before or after the data arrived? A preregistered primary outcome is a strong signal. A result that appears only in a subgroup, with no statement that the subgroup was planned, is a much weaker one.

None of those five needs you to know the field. They are the same habit of asking why a claim should be believed that powers elaborative interrogation as a study method, pointed at a paper instead of at a textbook fact.

How do you tell what the data shows from what the authors claim?

Write the finding in your own words before reading the discussion, then compare. The gap between your sentence and theirs is the paper's argument, and it is the thing worth examining.

Four gaps recur often enough to watch for by name:

The third gap deserves particular attention because it hides behind a technical word. Statistical significance is a statement about how surprising a result would be if nothing were going on. It is not a statement about how big the effect is. A large enough study can return a significant result for a difference far too small to change anything, so always find the effect size and ask whether a difference of that magnitude would matter to anyone.

Note also that the fourth gap is why the limitations paragraph is worth reading closely rather than politely. Authors are usually honest there. The claim in the abstract is simply broader than the claim they defend when being careful.

What to do when the jargon stops you

Do not look up every unfamiliar term as you meet it. That converts one hour of reading into four and breaks the thread of the argument each time. Use a two-tier rule instead.

First tier, keep going. Mark the term and read on. Most technical terms are defined implicitly by their second or third appearance, and a surprising number turn out to be labels for things that do not affect the argument at all: an assay name, a software package, a strain designation. If a term appears once in the methods and never again, it did not need to stop you.

Second tier, stop and resolve. A term earns a real lookup when the finding depends on it, when it appears in the figure axes, or when it turns up in the sentence stating the main result. Those are load bearing and no amount of context will fill them in.

Two habits make this faster over time. Keep one running glossary per field rather than per paper, since the same forty terms recur across everything you read in a subfield and the curve flattens quickly. And when a concept rather than a word is the blocker, go find a review article or a textbook treatment of it instead of rereading the same paragraph, which is the specific move that our guide to improving reading comprehension calls noticing the exact moment you stopped understanding. Rereading a sentence you lack the background for just produces the sentence again.

How to take notes on a paper you have to remember

Notes on a paper are not a shorter paper. If you are highlighting or transcribing sentences, you will finish with a compressed copy you understand no better than the original. Write answers to a fixed set of questions instead, so every paper you read produces the same shaped record and the records become comparable.

Six lines is enough for most papers: the question the paper asked, what was actually done in one sentence, the single most important figure and what it shows, the finding in your own words, the main limitation, and why this paper matters to you specifically. That last line does most of the work three weeks later when you are trying to remember why you saved it.

The discipline of compressing rather than copying is the same one behind summarising a textbook chapter properly, and it fails in the same way when you write with the source open in front of you. Close the paper before you write the six lines. What you can produce from memory is what you understood.

It is worth being clear that this is a different procedure from the structured reading systems built for textbooks. The SQ3R method for reading a textbook works by surveying, turning headings into questions, and reciting after each section, which fits a text written to teach you in sequence. A paper is not written to teach you and its headings are not a curriculum. Different document, different method.

Who needs this and when

Journal articles used to appear late, in a final year project or a graduate seminar. They now show up much earlier: in first-year lab reports, in evidence-based practice assignments for nursing and allied health, in psychology and economics courses built around replication, and across pre-med coursework.

The pressure is sharpest for pre-med students, since the reasoning tested by MCAT passages is close to what this post describes. Those passages present unfamiliar experimental setups with figures and ask what the data does and does not support, which is exactly the distinction between the results and the discussion. Practising on real papers builds the same muscle, and our guide to studying for the MCAT covers how that passage-based reasoning fits into a full study plan.

Two smaller points worth knowing regardless of field. Make sure what you have is a primary research paper and not a review: if there is no methods section describing something the authors did, it is a review, and it reports other people's findings filtered through an argument. Reviews are useful for orientation and are not evidence of anything on their own. And check the supplementary material when a paper matters, because the sample sizes, additional controls, and inconvenient panels often live there rather than in the main text.

How GeniusPal helps

The boundary first, because it is easy to overstate. GeniusPal will not read the paper for you. It does not summarise it, it does not extract its structure, it does not explain the methods, and there is no chat window to ask what a figure means. Judging what a paper shows is the entire skill, and handing that to software would leave you with the output and none of the ability.

What it does is test whether the reading stuck. Upload the paper itself as a PDF, or the six line notes you wrote about it as a text or Markdown file, and GeniusPal turns the material into a question set you can work through three ways: as a multiple-choice quiz, as flashcards, or in active recall mode where you write your answer from memory before seeing the model answer. Uploading your own notes is the more revealing of the two, because questions built from your summary catch the places your summary was vague.

That is a genuinely useful check for a seminar or a journal club, where the failure mode is arriving with a general impression of a paper and no ability to state its actual finding. The free plan covers 2 study sets for the lifetime of the account, which is enough to try it on one paper before deciding whether it earns a place in your routine.

The reading order itself costs nothing to try. Take the next paper you are assigned, open it at figure one instead of at the first paragraph, and write down what you think it found before you let the discussion tell you. The gap between those two sentences is the whole skill.