How to Read a Research Paper

If research papers feel dense or slow-going, you're not alone. Reading a paper in three passes makes it easier to digest. With each read, you go a little deeper than the last, so the big picture comes first, then the details. You can stop after any pass once you have what you need. As you read, keep one question in mind:

The three-pass approach comes from computer scientist Srinivasan Keshav, whose short paper "How to read a paper" (2007) has been adapted by researchers in many fields.

Tools like Consensus can help you find the best papers for your question. This guide covers what to do once you've found them.

Most scientific papers follow a similar structure. Each section answers a question, and knowing that question tells you what to look for.

Also glance at the authors, their affiliations, the funding source and any conflict of interest statement. Remember that the abstract is a summary written by the authors, not the evidence itself. It highlights what they want you to notice and leaves out most of the caveats.

The three passes work for any paper, but what to check first depends on the type.

New to a field?
Start with a review article. It summarizes what is known and points you to key studies.

Read with a pen or comment tool in hand. During Pass 1, jot down unfamiliar terms and keep going. From Pass 2 on, look them up or ask about them as soon as they block your understanding.

The sections later in this guide go with the passes: use "Reading tables and figures" in Pass 2, "Reading the Methods" in Passes 2 and 3, and "Significant is not the same as important" and the critical reading checklist in Pass 3.

By the end of your first pass, do three things.

In [who or what was studied], the researchers used [study design] to examine [question] and found [main result].

Read the Methods closely enough to understand what was done, and look carefully at the figures and tables. Then pick a key claim in the Results and find the numbers behind it in a table or figure. Do they support the claim as strongly as the sentence says? As you go, mark any references you want to read next.

Return to the question from the start: what would need to be true for me to believe this conclusion? Put yourself in the researchers' shoes. If you were doing this study, would you have made the same choices? Question each assumption and look for what is missing. Then make three separate judgments:

• Validity: Can I trust the result for the people or things studied?

• Importance: Is the effect large enough to matter, not just statistically significant? (See "Significant is not the same as important.")

• Applicability: Does it apply to my question, setting or population?

Copy this table for every paper you read and fill it in using your own words. If you can complete it without looking back, you've read the paper.

Here is how the three passes might go for an invented study on sleep and exam scores.

Pass 1. The abstract reports that students who slept longer before exams scored higher. First summary: In 400 first-year students at one university, the researchers used an observational study to examine whether sleep before exams is linked to exam scores and found that each extra hour of sleep went with scores about three points higher.

Pass 2. Table 2 shows the three-point difference, with a 95% confidence interval of 1 to 5 points. The interval does not include 0, so the result is unlikely to reflect no difference. The Discussion, however, says that sleep "improves" performance.

Pass 3. Validity: students chose how much to sleep, so a third factor such as stress or study habits could explain both their sleep and their scores. "Improves" is a causal claim that an observational design cannot support. Importance: three points on a 100-point exam is real but modest. Applicability: one university and one year group, so the finding may not apply elsewhere.

Revised claim
Among first-year students at one university, those who slept more before exams tended to score slightly higher.

The Methods is where most of your confidence in a paper should come from. Use these questions to work through it. You don't need to follow every statistical detail on a first read.

In Pass 2, use the first three questions to understand what the researchers did. In Pass 3, use the rest to question it.

Size reduces random error, but it does not fix a biased sample.

To see what a well-reported study of each design should include, check the EQUATOR Network reporting guidelines listed under Reading credits.

Use these checks in Pass 2. In Pass 1, read only the captions.

You don't need to master the statistics to read results. These four checks will take you a long way when it comes to reading quantitative studies.

For figures, read the axis labels and units before the data. Ask what the figure is comparing, then what pattern it shows.

Reading a qualitative study?
The checks above assume numbers. Instead, look at how participants were chosen, whether each theme is backed by quotes from participants, and whether the authors reflect on how their own perspective might shape the analysis.

A "statistically significant" result is unlikely to be due to chance. It is not necessarily large or meaningful. With a big enough sample, even a tiny difference can be significant.

Relative numbers can also make small effects sound dramatic. Suppose 2 in 100 people have a bad outcome without a treatment, and 1 in 100 with it:

• Relative reduction: 50%

• Absolute reduction: 1 percentage point (2% down to 1%)

Both are accurate descriptions of the same result. Whenever a paper reports a percentage change or "times more likely", look for the underlying counts.

Even good, peer-reviewed papers have limits. Spotting them means you are a critical reader; you are not attacking the authors. For checklists tailored to specific study designs, see the CASP checklists listed under Reading credits.

• Does the text match the data? Claims should be no stronger than the numbers behind them.

• Does the language fit the design? Words like "causes", "leads to" or "improves" need a design that can show cause and effect, usually an experiment.

• Is it a finding or a suggestion? Authors often offer possible explanations in the Discussion that their study did not test. Treat these as ideas, not facts. If you want to state one as fact, find a study that tested it.

• Are there small errors? Typos in tables or footnotes happen. Note them and check whether they affect the conclusions.

• Could something else explain the result? Look for other factors linked to both the exposure and the outcome that could account for what was found.

• Who collected the data, and how? Ask who was included, who was left out, and whether the method could bias results.

• Is the effect large enough to be meaningful? Check the size of the result, not just its significance.

• Where do the results apply? Note the place, population and time period. Findings may not apply beyond them.

• Do other studies agree? A single study rarely settles a question. Check whether other research reaches the same conclusion.

One paper is one piece of evidence. To judge how much weight it deserves, you need to know where it sits in the wider research. Consensus can help before, during and after you read.

Before you read: choose where to start

If you are new to a topic, search your question and filter the results by study type to find review articles and meta-analyses. A good review shows you what is already known, so you can tell whether a single study confirms it or challenges it.

While you read: use the Study Snapshot as a map

Each paper's Study Snapshot lists who was studied, how many, the methods and the outcomes. Use it to find your way into the Methods, then confirm each detail in the paper itself. If the Snapshot and the paper disagree, trust the paper.

After you read: check whether other studies agree

Turn the paper's main claim into a question and search it. Then ask:

• Do the results line up? For yes/no questions, the Consensus Meter shows how findings across studies split.

• Do stronger designs agree? Filter by study type. A finding from one observational study carries less weight if randomized trials found something different.

• Does it hold for other groups? Look at who was studied in other papers to judge whether the finding fits your question, setting or population.

What Consensus can't tell you
Consensus shows you what studies found. It can't tell you whether a particular study was well designed. That judgment comes from your own critical reading, using the checklist above.

An AI tool can explain and prompt, but you are responsible for your interpretation. Cite the paper, not the AI summary. Finding the passage a statement came from is not the same as reading the paper, so only cite papers you have read in full and can be sure support the statement you are making. For the full workflow, see How to Start a Literature Review in a New Field.

Reading credits

• Keshav, S. (2007). How to read a paper. ACM SIGCOMM Computer Communication Review, 37(3), 83–84. https://doi.org/10.1145/1273445.1273458. The original three-pass method this guide adapts, including the five Cs.

• Critical Appraisal Skills Programme (CASP). CASP checklists for appraising specific study designs. https://casp-uk.net/casp-tools-checklists/

• EQUATOR Network. Reporting guidelines that show what a well-reported study should include. https://www.equator-network.org/