Switch a filter unit on, hold a smoke pen against the edge of its gasket and watch where the smoke goes: some is drawn through the filter, and some slides sideways into the seam. In a test on 25 people, the particles crossing the seal leak of an N95 respirator outnumbered those passing through its filter medium by about an order of magnitude on average (Grinshpun et al., 2009). The rating on the box measures the medium. Nothing in the box measures the seam, or the rider it sits against.
Part of our Helmet air filter guide.
How do you test HEPA filter performance in a helmet? Run four checks: read the rating for the conditions it was graded under, run a smoke leak check around the gasket, run a two-sensor inside-versus-outside ratio test (effective efficiency = 1 minus inside divided by outside), and repeat that ratio after 30 days of use. The ratio test needs two identical sensors, and humidity, which is exactly what haze brings, can wreck it.
Key takeaways
- A filter's rating describes the medium, not the fit: on 25 people, particles through an N95's seal leak outnumbered those through its filter by about an order of magnitude (Grinshpun et al., 2009).
- Total inward leakage protection factors were under 2.3 for cloth and procedure masks, 6.2 for KN95 and 165.7 for N95, and the authors concluded that "mask performance is dominated by face seal leakage" (Duncan et al., 2021).
- Low-cost particle sensors read high above about 75% relative humidity (Jayaratne et al., 2018), and raw PurpleAir data overestimate PM2.5 by about 40% (Barkjohn et al., 2021), so compare two identical sensors rather than trusting one absolute number.
The research behind this
Peer-reviewed measurements this post is built on. Each card says what was measured, what was found, and what it cannot tell you.
Two pathways for particle penetration
- What they measured
- One N95 respirator and one surgical mask on 25 people doing fit-test exercises, then on breathing manikins: 5,250 penetration values for particles of 0.03 to 1 micrometre.
- What they found
- Particles through the seal leak far exceeded those through the filter medium: about an order of magnitude on average for the N95.
- What it can't tell you
- Health-care N95 and surgical mask, not a helmet gasket; the ratio belongs to that fit.
Filtration and total inward leakage of masks
- What they measured
- Cloth, procedure, KN95 and N95 masks on the bench at 17 L/min: how well the material stops particles, and how well the whole mask on a head does.
- What they found
- Material efficiency ran from 43.8% to 99.3%, but total inward leakage protection factors were under 2.3 for cloth and procedure masks, 6.2 for KN95 and 165.7 for N95.
- What it can't tell you
- Sedentary flow rate, masks not helmet-mounted filters.
Accuracy of fit checks versus fit tests
- What they measured
- Seven participants did a fit check, then a quantitative fit test on five N95s, a KN95, a surgical mask and fabric masks.
- What they found
- Fit-check responses correlated poorly with quantitative fit scores, and most N95s failed to fit the participants adequately.
- What it can't tell you
- Seven people and masks, not helmets. It supports "feel is not a test", not a helmet number.
Humidity and a low-cost particle sensor
- What they measured
- A Plantower PMS1003 low-cost sensor in the lab and in the field against humidity and fog, compared with a monitor with a dryer on its inlet.
- What they found
- Significant increases in reported particles above about 75% relative humidity; in fog the sensor's PM10 read 46% greater than the dried reference.
- What it can't tell you
- One sensor model, and the abstract gives no PM2.5 bias for haze.
A US-wide correction for PurpleAir PM2.5
- What they measured
- Almost 12,000 24-hour PM2.5 pairs from PurpleAir sensors and regulatory monitors across 16 US states, including smoke-impacted days.
- What they found
- Raw PurpleAir data overestimate PM2.5 by about 40%, and adding a relative-humidity term further reduces the bias.
- What it can't tell you
- US conditions and 24-hour averages, not tropical humidity or seconds-long readings inside a helmet.
Face masks on the streets of Ho Chi Minh City
- What they measured
- Total inward leakage of six mask types on manikins at the curbside of two busy roads in peak traffic, by mass, number and surface area.
- What they found
- N95, valved and carbon-layer masks achieved 60 to 80% efficiency; surgical and cloth masks 25 to 60%, "largely due to inappropriate mask fitting".
- What it can't tell you
- Manikins carry no facial-shape variability, and these are masks, not helmet filters.
Commuter exposure by transport mode
- What they measured
- Motorcycle, car, bus and MRT commuters on the same Taipei routes, measuring PM10, PM2.5 and PM1 en route.
- What they found
- Motorcyclists, on the shortest trips (28.4 minutes), breathed the highest concentrations: PM2.5 of 67.5 µg/m³. Idling at lights increased exposure.
- What it can't tell you
- Taipei in 2008, no haze, and not measured inside a helmet.
What the studies actually measured
Seven studies carry this post, and they answer three different questions: where particles get in, whether you can feel it, and whether the sensor you would use to check is honest.
On where particles get in, Grinshpun and colleagues tested one N95 respirator and one surgical mask on 25 people doing standard fit-test exercises, then on breathing manikins replaying the same breathing. That produced 5,250 penetration values for particles between 0.03 and 1 micrometre, each split into the seal-leak path and the filter-medium path (Grinshpun et al., 2009). Duncan, Bodurtha and Naqvi measured cloth, procedure, KN95 and N95 masks on the bench at 17 L/min, once for the material alone (43.8% to 99.3% efficiency) and once as a whole mask, where the total inward leakage protection factor was under 2.3 for cloth and procedure masks, 6.2 for KN95 and 165.7 for N95 (Duncan et al., 2021). Velasco and colleagues then took six mask types outdoors: manikins at the curbside of two busy Ho Chi Minh City roads at peak traffic, where N95, valved and carbon-layer masks reached 60 to 80% and surgical and cloth masks 25 to 60% (Velasco et al., 2022).
On whether you can feel it, O'Kelly and colleagues had seven participants do a fit check and then a quantitative fit test. The two barely agreed, and most N95s failed to fit adequately (O'Kelly et al., 2021).
On whether the sensor is honest, Jayaratne and colleagues found a low-cost Plantower sensor reporting significantly more particles above about 75% relative humidity, and in fog its PM10 read 46% greater than a reference monitor with a dryer (Jayaratne et al., 2018). Barkjohn and colleagues compared almost 12,000 paired 24-hour readings across 16 US states and found raw PurpleAir PM2.5 about 40% high, improved by a humidity term (Barkjohn et al., 2021). And Tsai and colleagues supply the rider context: on the same Taipei routes, motorcyclists breathed PM2.5 of 67.5 µg/m³, the highest of four commuting modes, with idling at traffic lights adding to it (Tsai et al., 2008).
Ride with filtered air
Filters haze PM2.5 on every ride
Easi Breezi mounts on your scooter or motorbike and feeds H11 HEPA-filtered air into your helmet. Bike-powered, no batteries, no charging.
Why doesn't a 95% rating tell you how well your helmet filter works?
Because the rating is a property of the filter medium, and what you breathe is the sum of two paths: air that passes through the medium and air that slips past it through the seal. If the seal path carries a lot, the medium's grade barely matters. Duncan's team put it bluntly: "mask performance is dominated by face seal leakage" (Duncan et al., 2021). For the medium's own conditions, see why a filter's 95% is a speed, not a property.
Two air paths into the same helmet
Air takes every route open to it. The rated path is only one of them.
You can write the two paths as one formula. If the medium passes a fraction Pf of particles and the leak path passes R times as many as the medium does, then total penetration is Pf × (1 + R), and effective efficiency is one minus that. Grinshpun's N95 result puts R at about 10 for that respirator on a face (Grinshpun et al., 2009). A medium rated 95% has Pf = 0.05, so total penetration is 0.05 × 11 = 0.55: a 95% filter behind a 10-to-1 leak is a system that is about 45% effective. This is an illustration using an N95-on-a-face leak ratio. A helmet gasket is a different leak path and nobody has published its ratio, which is why you should measure yours.
Duncan's numbers show the same shape from the other side. A total inward leakage protection factor of 6.2 for KN95 masks corresponds to about 84% (1 − 1/6.2, my arithmetic), even though the KN95 material alone tested at 99.3% (Duncan et al., 2021). Move the sliders below to see how quickly a good medium gives its efficiency away.
Installed-efficiency calculator (illustrative)
Pick the medium, then set how much the seal leaks compared with the filter. Formula: total penetration = Pf × (1 + R).
| R | Medium 95% | Medium 99.3% |
|---|---|---|
| 0 | 95% | 99% |
| 1 | 90% | 99% |
| 3 | 80% | 97% |
| 7 | 60% | 94% |
| 10 | 45% | 92% |
| 20 | 0% | 85% |
Illustration only. R = 10 is the N95-on-a-face average from Grinshpun et al. (2009); 99.3% is the KN95 material efficiency from Duncan et al. (2021). A helmet gasket's own ratio has not been published, so measure yours with the ratio test below.
Why check your filter now?
Because the smoke has not gone away. As of 8.30am on 21 September, 21 air quality monitoring stations across Malaysia recorded "Unhealthy" API readings, 11 of them in Sarawak: Serian at 168, Kapit at 166, Sri Aman at 160 and Nilai at 158 (Malay Mail, 2026). Riders who bought a filter in the middle of a bad season rarely get a calm moment to ask whether it works. Do these checks now, indoors, on a clear day, so you have a number before the next bad week. For the wider picture, the live haze map and AQI for riders shows current readings.
How do you test HEPA filter performance in a helmet? Four checks
Each check catches something the others miss, and none of them is worth trusting alone.
Check 1: Read the rating for its conditions
A rating is a number measured at a stated flow rate, on a stated particle size, under a stated standard. Find all three. A rating with no flow rate cannot be compared with anything, and a rating for medium efficiency says nothing about the seal. If a seller cannot tell you the conditions, treat the number as a headline, not a specification. Our earlier post on what a helmet filter's 95% actually measures walks through the flow-rate half of this.
Check 2: Run the smoke leak check
This is the qualitative check. With the helmet on a stand, never on a person, run the filter unit and aim a low-output smoke source, such as a smoke pen or a stick of incense, around the gasket perimeter. Watch for smoke crossing the seam where it should not. It finds gross bypass only, in the same way a user seal check does: NIOSH describes a seal check as "a quick procedure you conduct every time you wear a respirator to make sure you donned the respirator properly", and separates it from a fit test that verifies the wearer gets "the expected protection" (NIOSH, Fit Test FAQs). O'Kelly's data show why the distinction matters: felt fit and measured fit barely correlated (O'Kelly et al., 2021). A pass on this check is necessary, not sufficient.
Safety, please read. Do this indoors with a window cracked. Never do it during a live haze event, never with the helmet on a person, and never burn anything in a closed room with people in it.
Check 3: Run the two-sensor ratio test
This is the quantitative check, and it gives you the number the box cannot. You need two identical PM2.5 sensors and a steady, non-toxic source of particles that you can hold constant. Then:
- Put the two sensors side by side and power them up for 10 minutes in the same air. They should agree with each other. If they do not, you have a sensor problem before you have a filter problem.
- Place one outside the helmet and one inside the breathing zone, on the stand, with the filter unit running and nobody breathing into it.
- Log both readings once they settle, along with the room humidity if you can measure it.
- Calculate effective efficiency = 1 − (inside ÷ outside). Inside reads 55% of outside means about 45% effective, which is the calculator's 95%-medium, 10-to-1 example.
- Swap the sensors' positions and repeat. If the ratio flips, the sensors disagree; if it holds, the filter and seal are what you are measuring. (This step is my reasoning, not a published protocol.)
- Repeat with a fan blowing at the helmet front for the "moving" case, and once more stationary. Motorcyclists' exposure rises while idling at lights (Tsai et al., 2008), so stationary and at-speed are different tests.
The humidity trap. Cheap optical sensors count water as particles. The Plantower sensor Jayaratne tested reported significant increases above about 75% relative humidity, and in fog its PM10 read 46% greater than a dried reference (Jayaratne et al., 2018). Barkjohn found raw PurpleAir PM2.5 about 40% high across the US, needing a humidity term to correct it (Barkjohn et al., 2021). If the air you ride in is humid, an absolute reading from one sensor can mislead badly. A ratio of two identical sensors at the same humidity is more robust, because the humidity bias should hit both alike. That is reasoning, not a measured result: neither study tested a two-sensor ratio. The US EPA's February 2021 report on sensor testing starts from the same problem, noting significant variability in the data quality of air sensors, and scopes its targets to ambient, outdoor, fixed-site, non-regulatory use, not a moving helmet (US EPA, 2021). Keep the test in a dry, stable room, and treat the ratio as good to a rough percentage, not to the decimal.
Low readings also make the ratio coarse, since a ratio of two small numbers jumps with each digit. If you can, use a source that puts both sensors well above their floor.
Check 4: Repeat after 30 days of real use
Run the same test, same source, same humidity, after a month of riding. A filter that measured well when new and poorly when installed is a gasket problem. One that drops with time is a loading or fit-drift problem. This step has no published study behind it; treat it as a comparison protocol, not a finding. Our post on when to replace a motorcycle helmet filter covers how loading and airflow interact, and washable versus disposable filters covers what rinsing does to the medium.
What can these checks not tell you?
Three things. First, particles smaller than a PM2.5 sensor cares about: Velasco found ultrafine particles under 50 nanometres dominated the number concentration at a busy Ho Chi Minh City curb (Velasco et al., 2022), and a mass reading such as PM2.5 does not describe that fraction. Second, gases: a particle filter does not remove them, which is why exhaust odours can persist behind a clean particle reading. Third, the rider: the test uses a helmet on a stand, so it cannot capture facial shape, beard, glasses or how the helmet moves at speed. The ratio is a comparison tool for your setup, not a health guarantee.
What should you do with the number?
Write it down. Tape a card inside the helmet bag with the date, the ratio, both readings and the humidity. In the first hour of the next bad season you will know whether the filter is holding; without a baseline you will be guessing from a box. For the wider context on filters and helmets, the helmet air filtration guide is the place to start, and a helmet with an air filter, answered in nine questions covers the fit-and-shelf questions riders ask most.
Design the leak out, then measure it anyway
A seal is a dependency. Easi Breezi does not ask a face seal to hold: a fan-driven unit sends filtered air to your helmet through a hose, via an H11 filter rated at 95% or more on PM2.5, powered from the vehicle. A delivered-air design still has leak paths, which is why these checks apply to it too, and why we would rather you measure than trust a rating, ours included. We do not publish a measured installed efficiency for it, because we have not measured one.
Frequently asked questions
How do I test a HEPA filter at home?
Use two identical PM2.5 sensors, one on each side of the filter, and a steady source you control. Compare inside to outside, and calculate effective efficiency as 1 − (inside ÷ outside). Confirm the sensors agree side by side first, because low-cost sensors read high in humid air (Jayaratne et al., 2018).
Can I test a helmet filter with a smoke stick?
A smoke stick finds gross bypass around the gasket, and only that. It works like a user seal check, which NIOSH separates from a fit test that verifies "the expected protection" (NIOSH, Fit Test FAQs). Run it indoors, on a stand, never on a person.
Do cheap PM2.5 sensors work in humid haze?
They work as comparison tools, not as absolute meters. One tested sensor's PM10 read 46% greater than a dried reference in fog (Jayaratne et al., 2018), and raw PurpleAir PM2.5 read about 40% high across the US (Barkjohn et al., 2021). Two identical sensors at the same humidity give a more robust ratio.
Is a helmet filter rated 95% really 95% efficient?
The medium may be; the system depends on the leak path. In one worked illustration, a 95% medium behind a 10-to-1 leak is about 45% effective overall, using an N95-on-a-face ratio (Grinshpun et al., 2009). A helmet gasket's own ratio has not been published.
How often should I retest?
Once now to set a baseline, then after 30 days of real use, at the same humidity and with the same source. Retest whenever you change the filter, the gasket or the helmet.
Ready to breathe cleaner on every ride?
Easi Breezi is $199, with worldwide shipping at a flat $15, duties and import taxes included. See the Easi Breezi clip-on helmet air purifier and buy yours, then run the ratio test on it and tell us your number.
Sources
- Grinshpun SA, Haruta H, Eninger RM, Reponen T, McKay RT, Lee SA (2009). Performance of an N95 filtering facepiece particulate respirator and a surgical mask during human breathing: two pathways for particle penetration. Journal of Occupational and Environmental Hygiene. doi:10.1080/15459620903120086
- Duncan S, Bodurtha P, Naqvi S (2021). The protective performance of reusable cloth face masks, disposable procedure masks, KN95 masks and N95 respirators: filtration and total inward leakage. PLOS ONE. doi:10.1371/journal.pone.0258191
- Jayaratne R, et al. (2018). The influence of humidity on the performance of a low-cost air particle mass sensor and the effect of atmospheric fog. Atmospheric Measurement Techniques, 11, 4883-4890. doi:10.5194/amt-11-4883-2018
- Barkjohn KK, Gantt B, Clements AL (2021). Development and application of a United States-wide correction for PM2.5 data collected with the PurpleAir sensor. Atmospheric Measurement Techniques, 14, 4617. doi:10.5194/amt-14-4617-2021
- O'Kelly E, Arora A, Pirog S, Ward J, Clarkson PJ (2021). Comparing the fit of N95, KN95, surgical, and cloth face masks and assessing the accuracy of fit checking. PLOS ONE. doi:10.1371/journal.pone.0245688
- Velasco E, Ha HH, Pham AD, Rastan S (2022). Effectiveness of wearing face masks against traffic particles on the streets of Ho Chi Minh City, Vietnam. Environmental Science: Atmospheres, 2, 1450-1468. doi:10.1039/d2ea00071g
- Tsai DH, Wu YH, Chan CC (2008). Comparisons of commuter's exposure to particulate matters while using different transportation modes. Science of the Total Environment. doi:10.1016/j.scitotenv.2008.06.016
- Malay Mail (2026-09-21). Haze update: 21 areas across Malaysia at "Unhealthy" pollution levels, reporting Department of Environment Malaysia API readings as of 8.30am. malaymail.com
- US NIOSH, National Personal Protective Technology Laboratory. Fit Test FAQs (CDC mirror). restoredcdc.org
- US EPA, Office of Research and Development (2021). Performance Testing Protocols, Metrics, and Target Values for Fine Particulate Matter Air Sensors. EPA Science Inventory
Written by Ash — mechanical engineer and founder of Easi Breezi, building an active HEPA filtration system for motorcycle helmets (patent pending). Based between Hong Kong and Bali, riding daily in the traffic this blog writes about.