Methodology
What’s Detected? presents official USDA Pesticide Data Program (PDP) records from annual releases 1994–2024. We are independent of USDA and EPA.
Where the data comes from
We use the USDA annual data archives and reference tables. The current release contains 335,940 samples and 52,898,231 original analytical results. It was reconstructed on September 25, 2026 from the preserved pilot and official archives. Original files and checksums are retained.
What is a sample?
A sample is a USDA laboratory sample and can contain multiple food units. A count of 500 samples does not mean 500 individual strawberries. Each sample is counted once in a food’s headline rate.
How we count detections
We retain USDA’s laboratory outcome. O, A and R are detections. N, NA, NR, ND and NP count as tested nondetections under the documented source definitions. A result below its listed detection limit retains the laboratory outcome and source annotation.
Actual-tested denominators
A food’s headline percentage counts samples with at least one detected analyte, divided by samples with at least one validated analyte test. A food–analyte percentage counts detected samples divided only by samples actually tested for that analyte. Those tested counts can differ for the same food.
Unvalidated NU results
104,685 results have the USDA code NU, “Non-Detect: Unvalidated Residue.” We preserve these records but exclude them from validated test counts and percentages. The source contains 807 analytes and 42,194 food–analyte pairs; 806 analytes and 42,160 pairs have validated tests. An unvalidated-only relationship has no validated detection rate.
Not detected is different from not tested
Not detected means a sample was tested without a validated detection. It does not establish absolute absence. Not tested or not covered means there is no validated test in this release. Missing rows and years are never converted to zero.
Years and changing tests
Charts use annual-release years; the database also preserves original collection dates. A few records were collected in a different calendar year from their annual file. Food selection, tested analytes, methods and detection limits change. We connect only consecutive tested years and leave gaps visible. Pooled results describe these historical records, not present-day prevalence or causal trends.
Separate food identities
Fresh, frozen, canned and grain commodity codes remain separate. Rolled oats, steel-cut oats, oat bran and oat grain are distinct. The targeted 2009 organic-lettuce study remains separate. Historical canned corn and peas use the linked sample’s corrected commodity identity, with the original result codes preserved according to USDA’s documented 2017 revisions.
What counts as an analyte?
Analytes include individual compounds, metabolites and isomers. We do not treat analytical test classes as pesticide-use categories or combine related analytes into pesticide-product counts. Counts may differ from USDA charts that combine parent compounds and metabolites.
Glyphosate coverage
This release has glyphosate tests for corn grain and soybean grain. It has no glyphosate tests for oats. A lack of tests cannot support a claim that USDA found no glyphosate in oats.
Which pages are published?
The reconstructed editorial screen requires food pages to have at least 500 validated-tested samples and two tested years. Analyte pages require at least 1,000 tests, two foods, two years and ten detections. Pair pages require at least 100 actual-tested samples, two years, ten detections and eligible parent pages. These are editorial rules, not USDA standards or proof of statistical representativeness. Sparse and zero-detection relationships remain in the eligible food’s detailed results without individual indexable pages.
What we do not conclude
Detection alone does not establish a health risk. We do not rank foods as safe, unsafe, clean or dirty. Organic-versus-conventional comparisons and EPA tolerance comparisons are deferred. No organic claim is not the same as conventional. Origin shares do not compare countries’ residue levels.
Verification and limitations
The reconstruction passed 339 checks, including 13 comparisons with USDA annual reports, all annual commodity sample-count checks, file checksums, raw-row reconstruction checks and database integrity. Preserved source anomalies include some missing reference labels, missing origin details and below-LOD detections. Tests validate the reconstruction; they do not remove the limits of the original sampling program.