Reviewers already know your study has weaknesses. What they want to see is whether you know which ones matter and what each does to your conclusions.
What reviewers look for in a limitations section
Every study has weaknesses, and a careful reviewer may have spotted several of yours before reaching the Discussion. When they get to the limitations, they are checking three things: whether the weaknesses they found are there, whether you understand what each one does to your results, and whether your conclusions have been sized to fit.
A paragraph that lists "retrospective design, small sample, single centre" fails all three. It names categories rather than problems, says nothing about consequences, and usually sits beside conclusions written as if none of it mattered. As we noted in 5 things reviewers wish authors knew, a limitation earns its place when it says what it prevents you from concluding.
The standards ask for more than a list. The ICMJE Recommendations tell authors to state the limitations of the study in the Discussion and to note important limitations in the abstract. The most useful single sentence comes from the STROBE checklist for observational studies, item 19: discuss limitations "taking into account sources of potential bias or imprecision," and discuss "both direction and magnitude of any potential bias." That is a good standard for any design.
It is also a standard authors often miss. The STROBE explanation and elaboration document cites a survey in which authors reported important weaknesses of their studies in a questionnaire, weaknesses that were missing from their published articles.
Choose the limitations that could change a conclusion
Start with a longer list than you will use. Go back through your methods and note every point where the study fell short of its ideal: how participants were selected and who declined, how exposures and outcomes were measured, confounders you could not measure, missing data and loss to follow-up, protocol deviations, the number of comparisons made, and the population and setting.
Then sort each candidate with one question: could this plausibly change the direction, size, precision or applicability of a main finding?
- Yes, for a primary outcome. It gets its own sentences, using the four-part structure below. The most serious one goes first.
- Yes, but only for a secondary or exploratory result. One sentence, tied to that result.
- No, or it applies to almost every study of this design. Leave it out, or fold it into the point it actually affects.
Ross and Bibler Zaidi, writing in Perspectives on Medical Education, make the same point: address the most salient limitations of your specific study, not general limitations of most studies, and avoid reducing them to stock themes such as a single institution, self-reported data or a small sample. Those can be real limitations. They become useful only when you say what they did.
A quick test for gaps: predict the first comment a methodologist would write about your study. If you can predict it, it belongs on the list.
Four parts for every important limitation
Write each limitation that survives the sort in four short parts. This builds on the elements Ross and Bibler Zaidi recommend (describe the limitation, explain its implication, give possible alternative approaches, describe steps taken to mitigate it) and adds STROBE's direction and magnitude.
- Name it precisely. What was missing, mismeasured or unrepresentative, in whom, and how much. "Smoking status was missing for [proportion] of participants" beats "some data were missing."
- Say which way it could move the result, and how far. Towards or away from no effect, or narrowing the patients and settings the findings apply to. If you cannot tell, say so and explain why.
- Say what you did about it. Adjustment, a sensitivity analysis, a comparison of participants with and without complete data, or nothing, if nothing was possible.
- Say what it means for the conclusion. Which finding still stands, which needs caution, and for whom.
A worked rewrite
Take a hypothetical retrospective cohort study comparing hospital readmission in patients who were and were not dispensed a drug after discharge. A typical first draft reads:
This study has several limitations. It was retrospective and may be subject to bias. The sample was small, and it was conducted at a single centre, which may limit generalisability.
The same study, with its limitations rewritten:
Exposure was defined from dispensing records, so some patients classed as exposed may not have taken the drug. If this misclassification was similar in both groups, it would tend to bias the estimate towards no association, and the true association may be stronger than reported. Frailty was not recorded. If frail patients were less likely to be prescribed the drug and more likely to be readmitted, part of the apparent benefit may reflect healthier patients receiving it. [Report the sensitivity analysis or E-value.] All patients were treated at one tertiary centre, so the findings may not apply to [settings whose case mix or follow-up care differs, and how].
The second version is longer, but every sentence gives a reviewer something to check, and none of it is defensive. The small sample has not been hidden. It belongs where it has a consequence: in the width of the confidence intervals, which you then discuss as imprecision.
Say which way the bias pushes
Direction is the easiest part to leave out, and it is where your knowledge of the study counts most. Answer the question that matches each limitation:
| Limitation | Question to answer in the text |
|---|---|
| Exposure measured imprecisely | Was the error similar across groups? STROBE notes this tends to bias towards no effect, but adjusted estimates can move either way when correlated risk factors are measured with different precision |
| Unmeasured or poorly measured confounder | How is it likely related to both exposure and outcome, and would that exaggerate or mask the association? |
| Loss to follow-up, non-response or missing data | Did the people you lost differ from those you kept, in baseline characteristics or likely outcome? |
| Non-adherence or crossover in a trial | Could the intention-to-treat estimate understate the effect of actually receiving the intervention? |
| Many outcomes, time points or subgroups | Which findings are exploratory, given that some significant results are likely to arise by chance? |
| Wide confidence interval | Which clinically important effects can your data not rule out? |
The CONSORT 2025 explanation and elaboration adds two cautions for that last row: do not interpret a non-significant result as showing that interventions are equivalent, and use the confidence interval to show whether the result is compatible with a clinically important effect, regardless of the P value. If your main result is null, our post on why negative and null results matter covers how to frame what the data can exclude.
Quantify magnitude where you can instead of asserting it. For unmeasured confounding in an observational study, one option is the E-value, which VanderWeele and Ding defined as the minimum strength of association, on the risk ratio scale, that an unmeasured confounder would need with both treatment and outcome to explain away the observed association. They suggest reporting it for the estimate and for the confidence limit closest to the null. The STROBE explanation adds a reminder worth keeping in view: the real range of uncertainty in an estimate is larger than the statistical uncertainty its confidence interval shows.
Match the section to your study design
The main reporting guidelines each ask for limitations, with different emphases. Check which guideline your target journal requires and use its current version.
| Design | Guideline item | What it asks you to cover |
|---|---|---|
| Randomised trial | CONSORT 2025, item 30 | Sources of potential bias, imprecision, generalisability and, if relevant, multiplicity of analyses |
| Observational study | STROBE, item 19 | Sources of potential bias or imprecision, with the direction and magnitude of any potential bias |
| Systematic review | PRISMA 2020, items 23b and 23c | Limitations of the evidence included in the review, and separately, limitations of the review processes used |
For trials, CONSORT 2025 supersedes CONSORT 2010, so work from the 2025 checklist rather than an older copy. Its explanation document also notes that internal validity is a prerequisite for external validity, because the results of a flawed trial are invalid. Deal with bias first, then discuss who the results apply to.
For systematic reviews, write two distinct paragraphs. One covers the evidence, for example risk of bias in the included studies or inconsistent results. The other covers your own process, for example restrictions on the databases or languages searched.
Strengths without spin
Limitations are often discussed alongside strengths, and some journals formalise the pairing. BMJ Open, for example, asks for a "Strengths and limitations of this study" section of no more than five bullet points relating to the methods rather than the results, published as a box after the abstract. The BMJ's suggested Discussion structure, which the CONSORT 2025 explanation recommends to trial authors, places the strengths and weaknesses of the study straight after the statement of principal findings.
Three rules keep strengths credible:
- Keep them methodological. Design, completeness of follow-up, prespecified analyses, validated measures. A finding is not a strength.
- Pair a strength with a limitation only when it genuinely offsets it. A large sample reduces imprecision. It does nothing about confounding.
- Drop priority claims. The ICMJE advises authors to avoid claiming priority, so "the first study to" does not belong here.
Replace defensive and generic wording
| Instead of | Write |
|---|---|
| "Our study has some limitations that should be considered." | The most important limitation, stated directly |
| "Results should be interpreted with caution." | Which result, and what the caution is |
| "This may limit generalisability." | The patients or settings the findings may not reach, and why |
| "Bias cannot be excluded." | The specific bias and the direction it would push |
| "Despite these limitations, our findings are robust." | The sensitivity analysis that supports that claim, or nothing |
| "Larger studies are needed." | The design or measurement that would resolve a named limitation |
Keep the limitations consistent with the rest of the paper. If the limitations say causal direction cannot be established, the abstract should not say the exposure "reduced" the outcome. And do not apologise: a limitation stated plainly reads as command of your study, while one wrapped in hedges reads as worry.
Placement, length and the abstract
Put limitations in the Discussion unless the journal says otherwise. The ICMJE guidance mentions them after placing findings in the context of other evidence, while The BMJ's structure puts strengths and weaknesses earlier; either order works when the journal does not specify. A subheading helps reviewers find the section, and the CONSORT 2025 explanation suggests subheadings in the Discussion. Carry the most important limitation into the abstract too, as the ICMJE asks. A single clause is often enough.
There is no required length: give each limitation the space its consequence deserves. If reviewers later raise a limitation you missed, add specific wording to the manuscript and quote it in your response letter; our guide to responding to peer review comments shows how.
Final check before you submit
- The limitation a methodologist would raise first is in the section, and it comes first
- Each important limitation says what happened, which way it could move the result, what you did about it, and what it means
- Stock points such as single centre, small sample or retrospective design either have a stated consequence or are gone
- Imprecision is discussed using confidence intervals, and a null result is not described as equivalence
- Subgroup, secondary and multiple comparisons are labelled as exploratory where appropriate
- Strengths are about methods, not results, and none claims priority
- The abstract notes the most important limitation
- The reporting guideline item for your design is addressed, using the current version
Rewrite the conclusion last
Once the limitations are honest and specific, reread your conclusion against them. Any claim that a limitation undermines needs narrowing, and doing that yourself is far better than waiting for a reviewer to ask.
Directive Publications lists what its reviewers assess, including the interpretation of results and the transparency of reporting, on its peer review page. For manuscript preparation and journal requirements, start with the resources for authors.
Frequently asked questions
Where does the limitations section go in a research paper?
In most journals it sits in the Discussion, either straight after the principal findings or after you compare your results with other evidence, depending on the structure the journal prefers. The ICMJE Recommendations also ask that the abstract note important limitations. Some journals want a separate strengths and limitations summary, such as a short bulleted box after the abstract, so check your target journal's author instructions.
How many limitations should a paper include?
There is no required number. Include every limitation that could plausibly change the direction, size, precision or applicability of a main finding, and give the most serious one the most space. Weaknesses that affect only a secondary result can share a sentence, and generic points that apply to almost every study of your design can usually be left out.
Does admitting limitations make rejection more likely?
Leaving a weakness out does not hide it, because reviewers are asked to judge your methods and an important flaw they find that you did not mention casts doubt on the rest of the paper. Reporting guidelines such as CONSORT and STROBE expect limitations to be discussed. If a limitation genuinely undermines your main conclusion, narrow the conclusion rather than removing the limitation.
What is the difference between a limitation and a delimitation?
A delimitation is a boundary you chose on purpose to keep the study focused, such as restricting eligibility to adults or to one care setting. A limitation is a weakness in design, conduct or data that could affect the validity or precision of the results. A delimitation is not a weakness in itself, but if a scope choice affects who your findings apply to, state that consequence in the limitations.
Should the limitations section suggest directions for future research?
Yes, when the suggestion is specific. The ICMJE asks authors to explore the implications of their findings for future research, and the most useful suggestion names the design, population or measurement that would resolve a limitation you have described. A generic call for larger studies adds nothing, so make it specific or leave it out.