Original censusPublic data, fully reproducible
“We’re FDA registered.” I measured what that sentence contains.
A supplier sends an FDA registration number, and the conversation often moves on too quickly. We analysed the openFDA registration-listing export dated 10 August 2026 and deduplicated every China record by establishment. 41,745 listing records resolve to 4,973 establishments. Registration is a filing, not an inspection. The useful evidence is what the filing says about the parties involved.
Retested on 6 September 2026 against the export dated 4 September 2026, with the same filter and the same de-duplication key. 42,098 listing records now resolve to 5,012 establishments. The median establishment still carries 4 listings and the largest carries 651. Twenty-five days added 353 listings and 39 establishments, and the shape held.
FDA CENSUS / 01
Registration is a filing. It is not an inspection, an approval, or a reference.
Everything below is descriptive. None of it says a company is good or bad, and a name that differs from another name is a structural observation rather than an accusation. The reason to publish it is that the words on a registration are routinely read as saying more than they do.
What I measured, and the denominator that traps people
The frame is every record in the openFDA device/registrationlisting bulk export whose registration country is China. No sampling inside the frame. The export used carries 333,181 records across all countries and is dated 10 August 2026.
China accounts for 41,745 listing records. Those records are device listings, not companies. Deduplicated on registration number they resolve to 4,973 establishments, a median of 4 listings each and one firm carrying 648.
This matters more than it sounds. Querying the same public API for a count of establishment types returns 14,368 records carrying the export-only role, a number that reads like fourteen thousand exporters. Per establishment the export-only figure is 170. Anyone quoting record counts as firm counts is inflating by roughly eight times, and the API will hand you the inflated number without warning you.
What the declared roles say
An establishment declares one or more roles at registration, so these shares sum above 100%.
| Declared role | Establishments | Share |
|---|---|---|
| Manufacture Medical Device | 4,193 | 84.3% |
| Manufacture Medical Device for Another Party (contract manufacturer) | 1,928 | 38.8% |
| Export to the United States but perform no other operation | 1,828 | 36.8% |
| Repack or relabel | 317 | 6.4% |
| Develop specifications but do not manufacture at this facility | 189 | 3.8% |
| Foreign private label distributor | 136 | 2.7% |
The second line is the one worth sitting with. Nearly four in ten of these establishments have told the regulator, in a filing, that they make devices for somebody else’s name. That is ordinary contract manufacturing and there is nothing wrong with it. It does mean the brand on the box, the establishment that made it and the entity that ships it are three questions, not one.
Only 170 establishments (3.4%) carry the export role and nothing else. The common picture of a Chinese “trading company” that only ships is, in this population, rare. The far more common picture is a factory that also exports, or a factory that also makes for others.
The same structural question from the Chinese side of the record: who is the exporter, and why business scope no longer answers it.
The US agent is a shared address, not a reference
A foreign establishment must name a US agent, so 100% of them have one by construction. The finding is not the presence. It is the concentration.
Those 4,973 establishments name 1,466 distinct US agents. The distribution is extremely uneven:
732 establishments (14.7%) name the same agent, Regrek LLC.
331 (6.7%) name Registrar Corp.
134 name IRC USA, and the tail continues.
The ten largest agents together cover 1,763 of the 4,973 establishments, 35.5%.
At the other end, 1,111 of the 1,466 agents (75.8%) serve exactly one establishment: often the firm’s own US entity or an individual.
For a buyer the practical reading is short. A US agent on file tells you the establishment met a filing requirement. It is not a due-diligence signal, it is not a business reference, and if you were treating “they have a US representative” as reassurance, that reassurance is shared with several hundred unrelated factories.
Name continuity between establishment and owner
Each registration carries both the establishment name and the owner-operator firm name. We compared them after case folding, stripping punctuation and removing common corporate suffixes, so that a formatting variant does not count as a difference.
561 of 4,973 establishments (11.3%) carry an owner-operator name that is genuinely a different name.
Many are unremarkable group structures, a China subsidiary held by a foreign parent, where the two names obviously belong together. Others are two unrelated-looking Chinese names. We are not asserting anything about either kind. The point is narrower: if you check “the company” by one of these names, roughly one time in nine you have checked a name the record does not treat as the same party. The registration itself will not flag that for you.
The same effect measured from the Chinese registry, where 53.4% of a manufacturer census carried at least one former name: the registration census.
What this study cannot show
Medical devices only. Nothing here generalises to furniture, electronics or anything else. It is one regulated category with an unusually good public record.
US-facing registrations only. A Chinese factory with no US device business does not appear at all, so this is not a picture of Chinese manufacturing.
Registration is not approval. It does not mean the facility was inspected, or that its products were cleared. Declared roles are self-declared.
A different name is not wrongdoing. We report the frequency of a structural pattern and deliberately do not interpret individual cases.
One snapshot. The 10 August 2026 export, with no time series behind it.
We did not contact any establishment, and we published no personal names. Every figure comes from a public bulk file that requires no account.
The data, and reproducing it
The aggregated dataset behind every number on this page: fda-china-device-registration-census-2026-08-12.json.
Archived copies, each with its own DOI, resolving independently of this site: Zenodo · Harvard Dataverse · figshare.
A citation with everything it needs:
Currawong, “What a China FDA registration tells you: 4,973 measured”, census of the openFDA device registration-listing bulk export of 10 August 2026, deduplicated to 4,973 establishments, released 12 August 2026.
https://currawongweb.com/verify/china-fda-registration-study/
BibTeX
@dataset{currawong_verify_china_fda_registration_study_2026,
author = {Bao L. Zhou},
title = {{Chinese medical-device establishments registered with the US FDA: roles, name continuity and US agent concentration}},
year = {2026},
publisher = {Zenodo},
doi = {10.5281/zenodo.21897194},
url = {https://doi.org/10.5281/zenodo.21897194}
}To reproduce it, download the two openFDA bulk partitions for device/registrationlisting, keep the records whose registration.iso_country_code is CN, deduplicate on registration_number, and recompute. No API key is required and no login is involved.
The web download is the current privacy-clean distribution: US-agent labels are replaced by anonymous ranks while counts and shares remain. The earlier 12 August archive predates that change and is therefore treated as source history, not the canonical copy of this release. Published under CC BY 4.0.
Browse all measured studies and methods in the research index.
Not legal advice, and not a rating of any company named or unnamed. A record check reports what a record says on the day it was read, the FDA file on the US side, and the GSXT record on the Chinese side. Chinese supplier verification means holding both: the English filing alone cannot verify Chinese company identity, and the Chinese record alone says nothing about the US listing.
How we checked
Interface coverage was measured by calling each documented endpoint ourselves and recording which ones returned data. The most recent query date on this page is 6 September 2026. Where a table carries its own date, that date governs.
Being pushed to pay a deposit right now? The checks that matter before money moves take about ten minutes and cost nothing.
If you want these records pulled for your own supplier: the “Just check who they are” selection of the report menu covers them, packs from $26.55. Delivery follows the window on your order confirmation. Buying from a Chinese seller of record is the other route entirely. Its own choice, its own trade-offs.