Short answer: Researchers can deliberately infect volunteers with influenza virus. Yet getting flu to pass from one infected person to another in a controlled experiment has sometimes proved unexpectedly difficult. Experiments in 1918, 2020 and 2026 all produced strikingly low transmission. The puzzle is not whether influenza spreads between people, but why some seemingly intense exposures produce no detectable infections while others do.

In late 1918, ten young sailors walked into a hospital ward in Chelsea, Massachusetts. The ward was full of patients diagnosed with influenza, the disease then sweeping across the world. The sailors had volunteered for a direct test of how it spread.

Each sailor sat beside a patient and talked with him for five minutes. The two then brought their faces to within roughly two inches of each other. The patient breathed forcefully towards the sailor five times and coughed directly into his face five more times. The sailor repeated the procedure with ten different patients.

All ten volunteers were observed for a week. Not one developed the illness the investigators were trying to reproduce.

More than a century later, that result is still worth looking at. Not because it overturns everything we know about influenza, but because a modern version of the same difficulty has appeared in controlled research.

The experiment that refused to work

The 1918 work was directed by physician Milton Rosenau and colleagues from the US Navy and Public Health Service. About a hundred volunteers participated in the wider programme at Gallops Island in Boston Harbour, although only ten took part in the hospital-ward encounter described above.

The investigators tried much more than face-to-face exposure. They sprayed material collected from ill patients into the noses, throats and eyes of healthy volunteers. They tried direct transfers from nose to nose and throat to throat. Other trials used filtered secretions, blood, or cultures of bacteria then suspected of causing influenza. The attempts described in Rosenau’s report did not reproduce clinical influenza in the volunteers.

The team did not simply ignore possible flaws. They shortened the time between collecting patients’ secretions and giving them to volunteers. They eliminated saline and transport delays in some procedures. Still, the illness did not appear.

Rosenau did not claim to have disproved contagion. He wondered whether the volunteers had already acquired immunity and whether crucial conditions had been missed. His conclusion included an unusually candid possibility: “Perhaps there are factors, or a factor, in the transmission of influenza that we do not know.”

There are important limits to the historical evidence. Influenza virus had not yet been isolated in humans. The diagnosis of donors was clinical, so researchers could not verify their virus or measure infectious shedding. They could not test volunteers for silent infection or quantify their pre-existing immunity. Seven days without symptoms was a meaningful observation, but it was not the same as today’s PCR and antibody monitoring.

Even allowing for all that, the scene is hard to forget: ten sailors, a hundred close encounters with symptomatic patients, and no clinical influenza.

A century later, the numbers were still surprising

In a 2012 proof-of-concept study, researchers first infected volunteers with influenza A and then asked other volunteers to spend time with them. Three of fifteen exposed participants developed laboratory-confirmed infection. Transmission was possible, but far from automatic.

A larger trial reported in 2020 produced a more awkward result. Fifty-two volunteers were inoculated to act as potential donors, and 75 recipients were exposed over four days. Only one recipient showed evidence of transmission, detected by an antibody response rather than a positive respiratory PCR test or influenza symptoms. The researchers had anticipated substantially more infections.

Perhaps the laboratory virus made the donors unusually poor at passing on infection. That was one interpretation. So another group later tried something different: donors who had caught ordinary seasonal influenza in the community.

The EMIT-2 study, published in January 2026, enrolled five people with confirmed natural influenza infections and eleven healthy recipients. The participants interacted in a controlled room with little fresh-air ventilation, although the room’s air was also rapidly mixed by recirculation.

The researchers checked respiratory samples, symptoms and blood for evidence of infection.

Result: zero confirmed infections among the eleven recipients.

That result was not what the investigators had designed the study to achieve. Nor did they interpret it as evidence that influenza cannot spread. Instead, they examined what their experiment might reveal about the conditions transmission requires.

But scientists can infect volunteers directly

Here is the apparent contradiction. In another study published in 2026, researchers delivered measured doses of aerosolised influenza A/H3N2 directly to healthy volunteers. At the higher tested concentrations, three of four participants became infected with one delivery system and two of four with another. The infections were generally mild, and investigators measured viral shedding afterwards.

These are small groups, but they establish an important distinction. Delivering an infectious aerosol to a person’s airways is not the same experiment as asking a sick person to generate the right exposure for someone else. The first controls the virus preparation and delivery. The second depends on a chain of less predictable events.

A virus may be capable of infection without every encounter with an infected person delivering an effective dose.

So why did the person-to-person experiments fail?

The possible answers are more interesting than a single excuse.

The donor may not release enough infectious virus. In EMIT-2, influenza RNA was found in some exhaled aerosol samples, but culturable virus was detected in only a small fraction. The donors also coughed infrequently. A positive nasal test tells us that someone is infected; it does not tell us exactly how much viable virus that person is putting into the air at a particular moment.

The recipient may not be equally susceptible. Many of the 2026 recipients had immune responses suggesting previous encounters with related influenza strains. Researchers in 1918 had no way to measure that. Prior immunity can alter the outcome even when exposure occurs.

Airflow may change what reaches the next person. The 2026 team deliberately limited fresh-air ventilation but had vigorous recirculation. That could have dispersed concentrated exhaled breath before it reached a recipient. A room with plenty of virus-containing air is not necessarily the same as a brief inhalation of a dense plume from another person’s breath.

Timing, duration and chance matter. Infectious shedding changes during illness, and one close conversation is not equivalent to living together for days. Zero cases among eleven people is a striking result, but it cannot establish a zero probability of transmission. The study was small enough to miss an event that happens only occasionally.

These explanations are plausible and partially supported by measurements. They are not a solved reconstruction of what happened in each failed experiment. That distinction matters.

What happens outside the laboratory?

Real households provide a useful counterweight. In a prospective US study published in 2026, researchers detected influenza in 89 of 941 household contacts within one to seven days of the first infected person in the home: an estimated secondary infection risk of 9.5%. Greater viral load in the first case was associated with a substantially higher risk of another infection.

Genetic sequencing of available virus pairs also supported a connection between many initial and subsequent cases. The study could not identify the exact moment or route of each infection, and people can share exposures outside the home. Nevertheless, the data are consistent with human-to-human transmission and show something else that often gets lost in simple accounts: most exposed household members did not become infected during the observed interval.

The household findings and the laboratory trials are not interchangeable. They involved different ages, seasons, strains, environments and methods. Their contrast is a reason to investigate those differences, not to average the numbers together and declare the problem solved.

The experiment we still need

A particularly informative transmission study would follow naturally infected donors and recipients together, while measuring infectious virus in exhaled air, not merely RNA in the nose. It would document prior immunity, sequence viruses from both people, measure local airflow and the duration of exposure, and recruit enough participants across different seasons to separate unusual luck from repeatable patterns.

Even then, the study might discover that there is no single missing ingredient. Perhaps a small number of highly infectious donors account for a disproportionate share of spread. Perhaps certain combinations of shedding, immunity and air movement are required. Or perhaps current experimental rooms change the exposure they were built to observe.

Each possibility makes a different prediction that could be tested.

The unanswered question

Rosenau’s sailors never developed the illness his team expected. In 2020, a carefully arranged trial recorded only one confirmed infection among 75 recipients. In 2026, researchers used naturally infected donors and still observed none among eleven recipients. Meanwhile, deliberate aerosol challenges can produce infection, and household surveillance documents secondary cases.

These findings do not cancel each other out. They point to a boundary we still cannot draw precisely: when does being near an infected person become an exposure that actually causes infection?

In 1919, Rosenau suspected that something essential remained unknown. More than a century of virology has answered enormous questions since then. This particular one still resists a simple formula.

Sources and methods

Reading the numbers: A “recipient” in a controlled exposure study is not equivalent to a household contact, and a symptom-free volunteer in 1918 was not tested with modern laboratory methods. These differences limit direct comparisons. This article discusses published research, not advice to deliberately expose anyone to infection.