How to read a research paper as a beginner

Do not read a research paper from the first line to the last. Read it in up to three passes: a five-to-ten-minute skim of the title, abstract, introduction, headings, and conclusions; a careful second read of up to an hour that focuses on the figures; and a deep third pass only when you need it. Then judge the evidence: what kind of study it is, how large it is, what its limitations are, and whether it has been peer reviewed.

The three-pass method

The method comes from S. Keshav's short paper "How to Read a Paper," published in ACM SIGCOMM Computer Communication Review in 2007. Keshav wrote it for graduate students in computer science, but the steps work in any field. His key idea: "read the paper in up to three passes, instead of starting at the beginning and plowing your way to the end."

Pass 1: the bird's-eye view (5 to 10 minutes)

  1. Carefully read the title, abstract, and introduction.
  2. Read the section and subsection headings, and ignore everything else.
  3. Read the conclusions.
  4. Glance over the references, and note the ones you already know.

At the end, you should be able to answer Keshav's "five Cs":

Many papers stop here. That is fine. Keshav says the first pass is enough for a paper outside your area that may someday prove relevant.

Pass 2: the content (up to an hour)

Read with more care, but skip details such as proofs. Take notes in the margins. Look closely at the figures and graphs. Keshav asks: "Are the axes properly labeled?" He also checks whether results are shown with error bars. Mark references you have not read for later.

After this pass, you should be able to summarize the main point of the paper, with its evidence, to someone else. If you still do not understand it, Keshav lists your options: set it aside, read background material and come back, or go on to the third pass.

Pass 3: the deep read (hours)

The third pass means trying to "virtually re-implement the paper": make the same assumptions as the authors and re-create the work in your head. This is how you find hidden assumptions, missing citations, and weak points in the methods. Keshav estimates four or five hours for a beginner and about an hour for an experienced reader. Most learners need this pass only for the few papers their work depends on.

Read the abstract with care

The abstract is a summary written by the authors, and authors have an interest in a positive result. A study of 72 randomized controlled trials whose primary outcome was not statistically significant found "spin" in the conclusions of 58% of the abstracts and in the titles of 18% of the papers. Spin here means reporting that makes the treatment look beneficial anyway, or that pulls attention away from the null result.

So treat the abstract as a map, not a verdict. In the second pass, check that its claims match the results section.

Five questions for the methods section

The methods section is where you learn how much the result is worth. You do not need to follow every statistical step. Ask:

  1. Who or what was studied, and how many? Small studies are less reliable. A review of neuroscience research found that low statistical power leads to overestimated effect sizes and low reproducibility.
  2. Was there a comparison group? Were people assigned to groups at random?
  3. What was the main outcome? Was it chosen before the study, and is it the one the abstract talks about?
  4. How long did it run? A short study can miss later effects.
  5. Who paid for it? Read the funding and conflict-of-interest statements.

Read the limitations

Most papers include a limitations paragraph, often near the end of the discussion. It is the authors' own list of what their study cannot show. Read it on the first or second pass. It tells you how far the result travels: to which people, which settings, and which kinds of claims.

Where the study sits in the evidence hierarchy

In health and medicine, study designs are often ranked as a pyramid. A Miami University library guide lists the usual order from strongest to weakest:

  1. Systematic reviews and meta-analyses
  2. Randomized controlled trials
  3. Cohort studies
  4. Case-control studies
  5. Case series and case reports
  6. Background information and expert opinion

The pyramid is a rule of thumb. The University of Colorado's health sciences library notes that several rating systems exist, that "some studies are more appropriate for answering particular research questions," and that the hierarchy is "a decision tool to be combined with clinical judgement." Study quality still matters within each level. Other fields, such as computer science or history, judge evidence in their own ways.

Preprints vs peer-reviewed papers

A preprint is a paper posted online before peer review. Preprint servers say so clearly. medRxiv states: "Preprints are preliminary reports of work that have not been certified by peer review." It adds that they "should not be relied on to guide clinical practice or health-related behavior." arXiv says that its moderation process "is not a peer-review process."

That does not make preprints bad. One study followed bioRxiv and medRxiv preprints posted early in the COVID-19 pandemic that were later published in journals. The conclusions in the abstract changed in 7.2% of the non-COVID papers and 17.2% of the COVID papers, and most of those changes did not change the paper's conclusions in a meaningful way. Note the limit: the study looked only at preprints that made it into a journal.

Peer review is not a guarantee either. A Cochrane review found "little empirical evidence" that editorial peer review ensures the quality of biomedical research, while its authors stress that the methods to study it are hard. Some published papers are later retracted. Check a paper's status in the Retraction Watch database, which Crossref made public in 2023.

Practical rule: treat a preprint as a strong lead, a single peer-reviewed study as evidence, and a good systematic review as the best summary you are likely to find.

A one-page note for every paper

Write this after your second pass. It takes five minutes and makes the paper usable later.

Where Gregator fits

Papers are much easier when you know the background of a field. Gregator, our product, builds that background as a course from a topic you type, with research across seven search providers, including arXiv and Wikipedia. Each Chapter shows how many sources it used and how many were classified as peer-reviewed, and inline footnotes let you open the source behind a claim and read it with the three passes above. Gregator does not fact-check, and a paper found on arXiv may not have been peer reviewed. If your goal is to search and screen the literature itself, tools built for that fit better; see our comparisons with Elicit and Consensus.

Frequently asked questions

How long does it take to read a research paper?

Keshav's estimates: five to ten minutes for the first pass, up to an hour for the second, and four or five hours for a beginner's third pass. Most papers only need the first or second pass.

Which part of a research paper should I read first?

The title, abstract, and introduction, then the headings and conclusions. Read the methods and figures only once you know the paper is worth your time.

What is the difference between a preprint and a peer-reviewed paper?

A preprint is posted before any formal peer review. A peer-reviewed paper was checked by other experts before a journal published it. Both can be wrong, but preprints have had less outside scrutiny.

How do I know if a study is good quality?

Look at the design, the sample size, whether there was a randomized comparison group, the limitations section, and who funded it. Then check whether other studies, or a systematic review, reach the same result.

Do I need to understand the statistics?

Not every detail. Look for the size of the effect, the uncertainty around it, such as error bars or confidence intervals, and the number of participants.

Sources