Methodology & Sources

How many counties are in the United States?

There are two defensible answers — 3,143 and 3,144 — and the difference is Connecticut. Through 2021, Connecticut was divided into 8 counties. Effective June 2022, the Census Bureau retired those counties as governmental and statistical units and adopted Connecticut’s 9 councils of governments — its planning regions — as county-equivalents (TIGER/Line 2022 onward). Counting the 9 planning regions yields 3,144 counties and county-equivalents; counting the 8 obsolete counties yields 3,143.

CountyAtlas uses the current basis — the 9 planning regions — so its national total is 3,144. Sources that still report 3,143 are using the pre-2022 Connecticut geography; both numbers are correct for their vintage, and neither is a typo. The count also includes county-equivalents that are not called counties: Louisiana’s parishes, Alaska’s boroughs and census areas, and the ~40 independent cities (chiefly in Virginia) that belong to no county. U.S. territories (Puerto Rico, Guam, and others) have their own county-equivalents and are counted separately, not in the 3,144.

Data sources & vintages

Every figure carries the vintage of its source. The current build uses:

All of these are freely usable public sources. The published dataset contains federal and Wikidata data only — no proprietary or license-restricted inputs — which is exactly what makes it freely redistributable.

Suppression: why some values are blank, not guessed

CountyAtlas never prints a survey estimate it can’t stand behind. For each one we compute the coefficient of variation (CV) from its margin of error — standard error = MOE ÷ 1.645, then CV = standard error ÷ estimate — and suppress any estimate whose CV reaches 0.30, the threshold in the Census Bureau’s Statistical Quality Standard F1-6.

The gate rarely fires — on the current data, 12 of the 3,144 counties suppress at least one field — and that is by design, not luck. For the headline figures we deliberately use sources engineered for small-area reliability: SAIPE for income and poverty, and PEP for population (see below). Those clear the bar almost everywhere. The suppressions that remain are almost entirely educational attainment — the one field drawn from raw 5-year ACS — in the smallest counties, where even a five-year sample runs thin.

But the rule is absolute. Where an estimate is too noisy to trust, the cell is shown as suppressed — never as a blank that reads like zero, and never as an invented value. A cell you can’t trust is shown as untrustworthy, not filled in. Sentinel codes returned by the source APIs (negative placeholders that mean “no data”) are treated as missing, not as real numbers. That the gate fires so seldom is a statement about the sources we chose, not a softening of the rule.

Why PEP, not ACS, for population

Two Census programs report county population, and they answer different questions. The Population Estimates Program (PEP) gives a single, timely estimate for a specific date (July 1, 2023). The American Community Survey gives a 5-year average (2019–2023) with a margin of error. For a plain “how many people live here” count, PEP is the more current and more direct answer, so it is the population shown throughout the atlas. We still use ACS where only it has the field — median age and educational attainment — and there the CV gate above applies.

ZIP → county: the postal answer, not the geometric one

A ZIP code is a U.S. Postal Service construct — a set of mail-delivery routes, not a shape on a map — so “what county is this ZIP in?” is a question about where the mail goes. CountyAtlas answers it with the HUD-USPS ZIP-County crosswalk, built from USPS delivery-point data, and reports each county’s share of the ZIP’s residential addresses.

This is deliberately not the Census ZIP Code Tabulation Area (ZCTA), a polygon the Census draws to approximate a ZIP. The two disagree: ZIP 10505 (Katonah, New York) has most of its addresses in Putnam County by the postal crosswalk, but its ZCTA is drawn entirely inside Westchester County. Tools built on the ZCTA report the wrong primary county; the postal crosswalk reports where the addresses actually are.

The principle: verified or omitted

One rule governs everything above: a rendered sentence exists only when a real field backs it. National totals are computed from the data, not typed in; a figure with no reliable source produces no claim rather than a plausible-looking guess; and the boundary basis (which Connecticut geography, which vintage) is stated, not assumed. That discipline is what makes the atlas — and its dataset — citable.

The county seats follow the same rule as the suppressed estimates. Wikidata records a seat for 3,052 of the 3,144 counties (97%). The remaining 92 — chiefly the independent cities and the District of Columbia, which have no separate seat, plus a few boroughs and planning regions — are left blank in both the dataset and the pages, never guessed. A missing seat reads as missing, the same way a suppressed estimate does.

The dataset

Every county’s core fields are published as a single documented CSV — 3,144 rows, federal and Wikidata data only. See the data page for the column dictionary, download, and version history.