How to read patent number formats across US, CN, JP, KR, EP and TW
**Checked: 2026-09-21.** Patent-office interfaces and field names can change, so recheck the live office source when a record matters.
A patent number is not just a string of digits. It normally carries at least three pieces of identity: the office, the document or application number, and sometimes a kind code. The same case may have an application number, a publication number and a grant number. A good conversion preserves the identity before it tries to retrieve a file.
Quick comparison
| Office | Real input example | What the example means | Normalized form | Looked up through | Main caution |
|---|---|---|---|---|---|
| US | `US 2024/0039765 A1` | pre-grant publication | `US20240039765A1` | USPTO ODP | Do not confuse an application serial with a publication |
| CN | `CN 104704658 A` | invention publication | `CN104704658A` | direct format or OPS | Design applications use a different shape |
| JP | `特開2017-165727` | Japanese publication | `JP2017165727A` | GPSS when type is unclear | Native year notation carries meaning |
| KR | `KR 10-2017-0012345` | Korean publication | `KR20170012345A` | direct format or GPSS | A registration number may keep the `10` prefix |
| EP | `EP 1 234 567 A1` | EPO publication | `EP1234567A1` | EPO OPS | An EP record is not an EUIPO design record |
| TW | `TW 112306341` | Taiwan application-like reference | `TW112306341` | GPSS | A bare number does not identify I, M or D by itself |
The normalized value is useful for searching, not as a replacement for the original citation. Keep both. A number such as `CN 308690548 S` can be recognized and still fail to download if the source does not have that version.
US: publication, grant, design and application serial
DownloadPatent accepts separators in common US publication and grant forms. `US 4,704,658 B2` becomes `US4704658B2`; `US 2024/0039765 A1` becomes `US20240039765A1`. A US design such as `US D739,282` becomes `USD739282`. The `D` is part of the identity, not decoration.
An application serial such as `17/123,456` is not the same thing as a US publication number. When the input context identifies a US application, look it up as an application, for example in the USPTO's Patent Center. The application record may give a publication or patent number, and the result still needs to be checked against the requested application. This is why copying a serial into a publication-number search can produce a misleading “not found” result even when the application exists.
CN: invention, utility, design and application shapes
DownloadPatent treats a CN application shape such as `CN 202011505170.7` differently from a publication. The year and type digit carry information; type `1` is an invention application, type `2` a utility model application and type `3` a design application. The concrete design application example used in the product is `CN 202330520742.7`.
Publication examples include `CN 104704658 A` for an invention publication and `CN 308690548 S` for a design publication. The `CN` prefix must stay attached to the number because it selects the office. A clearly recognized CN application number is looked up through EPO OPS or Taiwan's GPSS, depending on how the reference is classified. A format match alone does not mean that a publication number or downloadable PDF exists.
JP: native notation, years and ambiguous bare numbers
`特開2017-165727` is normalized as `JP2017165727A`. The parser also understands Japanese era-style references such as `特開昭11-297662`, and it keeps the publication kind when a Western-form reference already supplies one. A grant example such as `JP5571308B2` is not reduced to an untyped seven-digit string.
Japanese bare numbers are deliberately handled conservatively. A seven-digit number without a clear kind or native publication marker can represent more than one document identity, so the parser does not invent a suffix. When the application or design type cannot be inferred from the text, the number is routed through GPSS rather than guessed from its length.
KR: publication prefixes and registration numbers
`KR 10-2017-0012345` is a publication-style reference and normalizes to `KR20170012345A`; the `10` is not retained in that canonical publication form. A grant such as `KR10-0751994B1` keeps the `10` prefix and becomes `KR100751994B1`. That difference is intentional: publication and grant formats do not have the same identity rules.
The parser does not blindly reconstruct every Korean design or registration label. If a label omits the type information needed to distinguish 10, 20 or 30 series, it can fail closed instead of silently creating a different document number. KR application references are one of the four non-Taiwan offices that can be resolved through TIPO GPSS.
EP: kind codes and families
`EP 1 234 567 A1` becomes `EP1234567A1`. The normal EP forms use a seven- or eight-digit number with A1, A2, B1 or B2. A single A or B can be expanded where there is enough context, but an arbitrary eight-digit value is not automatically repaired. The leading zero rules are deliberately narrow because dropping a zero can change the document.
Espacenet shows EP applications and their patent families. A family lookup is also how possible English counterparts of non-English JP, CN, KR and TW documents are found. A family candidate is additional metadata for review; it is not a statement that two documents have identical claim scope or legal effect. EP coverage also does not imply coverage of EUIPO registered designs.
TW: I, M and D
Taiwan certificate-style numbers carry the strongest visual clue: `TWI606827` is an invention, `TWM565313` a utility model and `TWD237262` a design. Keep the `TW` prefix and the letter. Removing it turns three different document classes into the same ambiguous digit sequence. A publication-style number such as `TW 2024/02021 A` is normalized with its publication information, while `TW 112306341` is treated as an application-like reference.
Bare JP and TW numbers without a reliable type signal need a record lookup before they are classified, because their document kind cannot be decided safely from the characters alone.
Why the same number succeeds in one system and fails in another
DownloadPatent checks a number in three steps. First, it classifies the pasted text as native notation, a publication, an application or an unsupported line. Second, it looks the number up in a patent-office record that fits its office and type. Third, it checks each candidate before accepting it: the digits and kind code must match, the office must be the one requested, a neighboring family number is rejected, and only a documented check-digit variation is allowed for an application match. A response that merely contains similar digits is not accepted.
A response from a patent-office service is not the same as a verified match. A source can respond successfully while reporting that the request failed, or return a record for a related application. DownloadPatent treats “the source responded” and “the application resolved to a verified publication” as separate outcomes, and only the second one leads to a download.
How to check without DownloadPatent
1. Preserve the original text exactly, including `US`, `CN`, `JP`, `KR`, `EP` or `TW`, native labels and kind codes. 2. Decide whether you have a publication, grant, application or family reference. Do not send a US application serial to a publication-number form. 3. Use the office source that owns that identity: USPTO ODP or Patent Center for US, CNIPA (or TIPO GPSS and Espacenet) for CN, JPO for JP, KIPRIS or GPSS for KR, TIPO GPSS for TW, EPO OPS or Espacenet for EP, and WIPO PatentScope for PCT/WIPO references. 4. Compare the returned application and publication fields with the original. If the source returns only a nearby family member, keep it as a lead rather than replacing the original. 5. Record the exact document version and kind code before downloading. If the official source cannot resolve it, report the missing correspondence instead of silently changing the number.
DownloadPatent applies the same identity checks when it resolves a mixed list, then presents the available original documents and English counterparts for review.