What several mold samples tell you that a single sample cannot, from a 67-home study.
If you are choosing between room-by-room mold testing and a single sample, here is the short answer from the research: one sample can misread a house. In a study of 67 homes, single samples were right about 82% of the time on average. Taking three samples inside each home and requiring two of them to agree sorted 21 of 22 homes correctly.
In this study, several samples read together did better than one. Mold-free homes also looked more like their own outdoor sample, which makes an outdoor sample a useful reference. For how an illumenair visit is set up, see how it works.
The study
Bridget Hegarty, Annabelle Pan, Ulla Haverinen-Shaughnessy, Richard Shaughnessy and Jordan Peccia published “DNA Sequence-Based Approach for Classifying the Mold Status of Buildings” in Environmental Science & Technology in 2020. Local professional building inspectors sampled single-family homes in 11 US cities, 5 to 10 homes per city. Some homes had inspector-confirmed water damage with visible mold. The rest had no signs or known history of dampness, water damage, mold growth or mold odor.
In each house the inspectors swabbed dust that had settled out of the air on top of three indoor door frames, plus one swab from the outdoor side of a door frame. In the moldy homes they also swabbed the mold itself, up to five samples. In all, 288 samples from 67 homes passed quality checks. The fungi in each sample were identified by their DNA.
The team then trained a computer model (a method called random forest) to tell the two kinds of homes apart from that DNA, holding back part of the data each time to test the model on samples it had not seen.
What they found
Judged one sample at a time, the first model was right 82% of the time on average: 86% for samples from moldy homes and 70% for samples from mold-free homes. Using only the fungi that differed between the two groups raised the mold-free figure to 83%. The largest source of error was clean homes being called moldy.
21 of 22 homes classified correctly when each home was judged by its three indoor samples and at least two had to be right: 12 of 12 moldy homes and 9 of 10 mold-free homes. The authors describe this as about 95% accuracy.
Two more findings matter for anyone reading a mold report. Mold-free homes shared more of their fungal types with their own outdoor sample than moldy homes did. And in moldy homes, one type tended to dominate: the most common type made up a median 64% of a sample from a moldy home, against 32% in a mold-free home.
Why it would happen
Mold growing on a damp material sheds its own spores into the house, so the indoor mix drifts away from the outdoor mix and toward a few dominant types. The authors report lower variety in moldy homes, consistent with that.
One sample also carries chance: whatever happened to land on that door frame. Two or three samples agreeing is harder to get by accident. That point is our reading; the study did not test it directly.
Distance mattered little in this study. The influence of the mold on settled dust fell off only slightly with distance from the growth (R² = 0.05), which suggests the benefit came from having more than one sample, read together, rather than from finding one perfect spot, though the study did not test this.
The limits, stated plainly
These were easy cases: visible, inspector-confirmed mold against homes with no history of water at all. Hidden mold, which is the hard case, was not tested. The authors also note that the full water history of an older house may not be known.
This was DNA sequencing of settled dust, not the spore-trap air sample most mold inspections use. The results do not transfer number for number to any other method. The paper says three indoor door frames; it does not say each was in a different room.
The study never compared one sample with three in the same homes as an experiment. The 82% and the 21 of 22 come from the same paper, but setting them side by side is our comparison, and the 21 of 22 used the refined model.
The model was built and tested on this one set of homes. The authors say whether it works elsewhere depends on testing in other regions. Climate appeared to shape the fungal mix. The sampling plan listed Portland, Oregon, among its cities, but none of the climate zones in the published results matches western Oregon and Washington’s, so we do not claim Portland homes are in the final data.
What survives: in these homes, several samples read together separated moldy from clean homes better than one sample alone, and mold-free homes looked more like their own outdoor sample.
Fixing it
This study tested detection, not repair. A sample’s job is to point an inspection toward moisture, and the fix is still drying the material and stopping the water.
What gets lost in the mold-test marketing
Mold reports often lean on two names, Aspergillus and Penicillium. In this study, the overall amounts of neither genus differed between moldy and mold-free homes (p = 0.98 and 0.42). Particular types within each did, but a genus name alone did not separate the houses. A test that reports only genus names cannot show you that difference.
What the data supports
- Ask how many samples a mold test includes, and where they go.
- Insist on an outdoor sample from the same property, the same day.
- Sample more than one spot: the rooms you live in, plus anywhere that smells musty or has had water.
- Read the pattern across samples, not one number.
Bottom line
One mold sample can misread a house. In this study, three samples per home, two agreeing, got 21 of 22 homes right.
First step: before you book any mold test, ask how many samples it includes and whether one is outdoors.
Mold testing in Portland and Vancouver
illumenair does indoor air quality and mold testing in Portland and Vancouver homes. We measure the air and explain what we found.
971-363-5626 · info@illumenair.com · www.illumenair.com
Source
Hegarty B, Pan A, Haverinen-Shaughnessy U, Shaughnessy R, Peccia J. “DNA Sequence-Based Approach for Classifying the Mold Status of Buildings.” Environmental Science & Technology 2020;54:15968-15975. doi:10.1021/acs.est.0c03904. Read in full October 11, 2026, via the NSF Public Access Repository: https://par.nsf.gov/servlets/purl/10208598