Methodology
Last updated: September 2026
Every figure on this site is computed from the public datasets below by a script we publish with the site's source. Nothing is hand-entered.
1. Predicted radon levels (model estimates)
Source: National Radon Exposure Model, file ZIPCODE_Average_Radon_2001_2021.csv, license CC0 1.0. The model is a random forest trained on more than six million radon measurements from independent laboratories (2001-2021) using roughly 200 geological, meteorological, architectural and socioeconomic predictors. For each of 29,215 ZIP codes in the conterminous US it gives a predicted long-term mean radon concentration in pCi/L for three settings, each with a standard deviation:
- Basement - the highest and most conservative figure; we grade on it.
- Upstairs (above-ground floors).
- Screening - what a typical screening test placed per common practice would read.
These are estimates of an area average. They are not measurements of any home, and individual homes commonly differ from the area mean by several times. We label them "predicted (model estimate)" everywhere. Alaska and Hawaii are outside the model.
2. EPA radon zones
Source: EPA Map of Radon Zones spreadsheet (EPA 402/A-16/001), public domain. The 1993 map assigns each county Zone 1 (predicted average indoor screening level above 4 pCi/L), Zone 2 (2-4) or Zone 3 (below 2). The file has no FIPS codes, so we match counties by normalized name to Census 2020 counties; renamed or split Alaska census areas and Oglala Lakota County, SD are mapped explicitly, with successor areas inheriting the old zone. All 3,143 current counties received a zone; four EPA rows for geographies that no longer exist (Yellowstone National Park, MT; Bedford, Clifton Forge and South Boston cities, VA) are dropped. The EPA itself notes that homes with high radon exist in all three zones.
3. Geography: ZIP to county
We use the Census Bureau's 2020 ZCTA-to-county relationship file. Each ZIP is assigned to the one county containing the largest share of its land area (its "primary county"). 29,218 of the model's 29,231 ZIPs matched a 2020 ZCTA; 13 did not and are omitted. A ZIP that straddles counties is counted only once, in its primary county.
4. County and state figures
A county's predicted basement, upstairs and screening values are the mean of its ZIPs' predictions weighted by each ZIP's 2010 Census population (the most recent ZCTA population available without an API key), so the figure describes where people live. The county SD shown is the population-weighted mean of the ZIP SDs - a typical ZIP-level uncertainty, not a formal county interval - and we also show the min-max range of ZIP predictions. 3,085 counties have predictions; 58 have none (Alaska, Hawaii, and 23 small counties or independent cities where no ZIP has most of its land) and show the EPA zone only. State and national averages are population-weighted across counties: the national predicted basement average is 1.8 pCi/L.
5. Grades
Grades use the EPA's own thresholds applied to the predicted basement mean: High ≥ 4 pCi/L (EPA action level), Moderate 2 to 4 (EPA "consider fixing"), Low below 2. A grade describes the typical home in an area and is a guide to how urgently to test, nothing more.
6. Measured Minnesota tests
Minnesota county pages additionally show measured results (mean, median, percent of tests at or above 4 pCi/L, tests per year, 2016-2025) from the Minnesota Department of Health. These are real tests, kept visually separate from model predictions. They skew high because testing is voluntary and often basement-focused. No other state's measured data is included yet.
7. Which pages exist
Every state and county has a page. ZIP pages are generated only for ZIPs predicted at or above 2.5 pCi/L in the basement or that are the most populous ZIP in their county; all other ZIPs appear in their county's table and in the homepage lookup.
Known limitations
- Predictions are 2001-2021 long-term averages; they do not reflect any single season or recent construction.
- ZIP codes are postal routes, not neat areas; ZCTA approximations introduce edge errors.
- 2010 population weights are old; they affect weighting only, not the radon values.
- The EPA zone map is from 1993 and can disagree with the model for a given county. We show both.
- Mitigation cost is stated as an industry-typical range, not a local quote.
Found an error? Tell us.