Thanks to the reliable laws of physics that describe the relationship between the size of a particle and the amount of light it gives off when it gets scattered by a laser beam into the lens of a camera, Fischer was able to back-calculate the size of the smallest particle they could detect: half a micron. Knowing that, he was able to quickly write a bit of computer code that opened the video footage, tracked individuals particles from frame to frame, and quantified the number of detectable particles emitted. In the end, this produced a picture of how many particles built up in the box during about 35 seconds of talking. That was the control.

Next, he repeated the experiment wearing 14 different masks, including N95s—valved and unvalved—surgical masks, a bandana, a spandex-blend neck gaiter, and cotton masks of varying designs. Then his team, which included a handful of physics and engineering colleagues collaborating remotely, compared the ratio of droplets produced while wearing each one to the no-mask control and ranked each one accordingly. These results were published Friday in Science Advances.

By far, the mask that was best at blocking a speaker’s exhaled particles was the fitted, unvalved N95, for which “we did not detect any particles at all,” says Fischer. The surgical mask performed similarly well, blocking almost all detectable speech particles, followed by cotton masks that contained a layer of polypropylene. Most other cotton masks fell into the middle of the pack, along with valved N95 masks, which are designed to protect the user from inhaled environmental threats like wildfire smoke, pollution, and pathogens—but because they contain an exhalation valve, do little to block potentially infectious particles from escaping. The bandana did next to nothing. But that wasn’t even the worst. The neck gaiter, made out of a lightweight, breathable fabric favored by runners and cyclists, let through even more particles than the control group—110 percent relative to wearing no mask at all.

If you’re wondering how that is even possible, you’re not alone. Fischer was similarly stumped. Then he went back and looked at the footage again of himself wearing the neck gaiter. “You can see that it’s not just that there are more particles, but that on average, the particles are much smaller,” he says. His team believes the stretchy, porous material is actually fracturing bigger, heavier droplets, splintering them into tinier particles that can more easily remain suspended in the air.

If that’s true, it would blow up the maxim that any mask is better than no mask, says Kimberly Prather, an environmental aerosol researcher at UC San Diego who was not involved in the study. But there’s another possible explanation: Maybe the extra particles aren’t all respiratory droplets. Instead, they could be fibers shedding off the material itself. This has been shown to happen before, and would be easy enough to test—but Fischer and his coauthors didn’t. “Splintering would be bad, but we don’t know for sure that’s what’s going on,” says Prather.

She also points out that the sample size for most of the mask testing is precisely one person. The study doesn’t capture all the variability in how people’s face shapes and speaking patterns might affect the effectiveness of different kinds of masks. So, while this project’s results are in line with other, larger, more rigorous studies, one shouldn’t read too much into the performance outcomes of individual masks based on this study alone, she says.

Still, Prather is impressed that the Duke team’s technique can detect particles down to half a micron. Most laser visualization methods are sensitive only to about 20 microns. “That’s a big deal, because this captures aerosols—the particles that come out during speech—not just bigger droplets emitted during coughing or sneezing,” she says. “Keeping it in perspective, I think it’ll be a great comparison tool to look at variability between people, more conditions. There’s a lot of different things you can do with the setup they’ve developed.”