What the evidence says about masks, and what it leaves open

People often ask a simple question: do masks work? The honest answer depends on which mask, worn by whom, in what setting, and measured how. This sheet walks through one widely discussed review, the 2023 Cochrane update on physical measures against respiratory viruses, and sets it beside three other opened sources so that the reader can see what each one does and does not say.

As at 30 January 2023 (Cochrane): the review was published online on that date. Findings below are as the review states them, and later studies may have changed the picture.

What the Cochrane review looked at

The review, titled Physical interventions to interrupt or reduce the spread of respiratory viruses, collects randomised controlled trials, the type of study in which people are assigned by chance to one approach or another. According to its Cochrane Library page, it identified 78 relevant studies, run in hospitals, schools, homes, offices, childcare centres and communities, in both epidemic and non-epidemic periods. It assessed medical or surgical masks, N95/P2 respirators and hand hygiene.

For masks worn in the community compared with no mask, the review reports that wearing a mask "may make little to no difference" in how many people caught a flu-like or COVID-like illness (9 studies, 276,917 people), and "probably makes little or no difference" to laboratory-confirmed flu or COVID (6 studies, 13,919 people). For N95/P2 respirators compared with medical or surgical masks, it reports that they probably make little to no difference to confirmed flu (5 studies, 8,407 people; moderate-certainty evidence).

What the authors say about their own certainty

The summary's key message is a statement of uncertainty: "We are uncertain whether wearing masks or N95/P2 respirators helps to slow the spread of respiratory viruses based on the studies we assessed." The authors add that the trials had a high risk of bias, that outcomes were measured in different ways, and that adherence to the interventions was relatively low, all of which "hampers drawing firm conclusions."

Uncertainty is a finding about the quality of the evidence. It differs from a measured result of zero benefit, and it also differs from proof of benefit. The authors themselves write that "the true effect may be different from the observed estimate" and call for large, well-designed trials that measure adherence. Readers who want the review's own wording on how it was received should read it on the Cochrane Library; the editors could not open Cochrane's separate editorial statement on the review, so this sheet does not summarise it.

Why mask trials are hard to run and read

Adherence is the first problem. A person assigned to wear a mask may wear it for part of the day, pull it down to talk, or stop after a week. The review treats low adherence as a core limit, because a trial then compares "assigned to masks" with "assigned to no masks" rather than real mask use with none.

Fit is the second. The CDC mask page (updated 18 August 2025) says that how well masks block viruses depends on design and on how well they fit, and that gaps let air leak in and out. A trial that hands out one mask type to thousands of faces gets a wide spread of fits. See fit and seal for how a leak path forms.

What is measured is the third. The review distinguishes symptom-based outcomes such as flu-like illness from laboratory-confirmed infection, and notes that the former are subjective. The same CDC page describes two jobs for a mask: reducing what an infected wearer releases, and reducing what a wearer breathes in. A trial that counts only the wearer's own infections speaks mainly to the second job. That reading is the editors' observation, not a statement from the review.

Respirators versus medical masks in health care

Two further sources look at health-care workers. A systematic review in Influenza and Other Respiratory Viruses (published 21 April 2020) pooled four trials and found that medical masks did not increase laboratory-confirmed viral respiratory infection compared with N95 respirators (odds ratio 1.06, 95% confidence interval 0.90 to 1.25), with low certainty. Its authors concluded the two offered similar protection during non-aerosol-generating care, and its introduction notes a consensus that N95 respirators protect better during aerosol-generating procedures such as intubation.

A randomised trial in the Annals of Internal Medicine (published 29 November 2022) enrolled 1,009 health-care workers at 29 facilities in four countries, comparing medical masks with fit-tested N95 respirators over 10 weeks. PCR-confirmed COVID-19 occurred in 10.46% of the medical mask group and 9.27% of the N95 group (hazard ratio 1.14, 95% confidence interval 0.77 to 1.69). The authors said the overall estimates rule out a doubling of hazard for medical masks, and also noted that results varied by country.

How to read it without overclaiming

Nothing above shows that masks are useless, and nothing above shows that every mask protects. A WHO page (last updated December 2021) frames masks as one part of a wider set of measures, and says a mask alone is not sufficient protection against COVID-19. The CDC page above adds that cloth masks generally protect wearers least and NIOSH Approved respirators such as N95 most, which matches the layout of the types sheet and the filtration ideas in how masks work.

This is general information. A person's own clinician or occupational-health adviser has the final word on what suits their circumstances.

The short version

The 2023 Cochrane review found low to moderate certainty evidence and said it could not draw firm conclusions about masks, partly because of low adherence, mixed outcome measures and bias risk in the trials. Studies comparing medical masks with N95 respirators in health-care workers found similar results for the outcomes measured, with caveats. Fit, adherence and what is counted all shape what a trial can show.