Every physical good imported into the United States gets a code from the Harmonized Tariff Schedule of the United States (HTSUS). The first six digits are international (the World Customs Organization's Harmonized System); the U.S. adds two more for the legal rate line, and two statistical digits on the end. Ten digits total. The eight-digit line is where the duty rate lives — the last two are for reporting, and usually carry a blank rate.
The code is not a matter of opinion. Classification is governed by six General Rules of Interpretation (the GRIs), applied in order. You do not skip to GRI 3 because it gives a nicer answer; you only reach it if GRI 1 and 2 don't resolve the good. That "in sequence" discipline is the whole game, and it's what separates a defensible code from a guess.
The six rules, in the order you apply them
GRI 1 — Classify by the terms of the headings and any relative Section or Chapter notes. Most goods stop here. The heading text and the legal notes are controlling; the titles of sections and chapters are for reference only. If the heading says "electrical machines and apparatus… having individual functions," and your good is exactly that, you're done at GRI 1.
GRI 2 — Extends headings to (a) incomplete or unassembled articles that have the essential character of the finished good, and (b) mixtures and combinations of a material or substance. This is what drags a composite good — plastic + electronics + textile — into a single-heading fight.
GRI 3 — The tie-breaker when a good is prima facie classifiable under two or more headings. It has three sub-rules, applied in order: 3(a) the most specific description wins; 3(b) failing that, classify by the material or component that gives the good its essential character; 3(c) failing even that, the heading that appears last in numerical order.
GRI 4 — Goods that fit nothing else go to the heading for the most similar goods. Rare.
GRI 5 — Cases, containers, and packing materials — when the case is specially fitted and sold with the article (a camera case with the camera), it follows the article.
GRI 6 — Repeats the same logic at the subheading level: to pick between two eight-digit lines under the same heading, you compare subheadings of the same level using GRI 1–5.
The same product, two codes, a 7% swing
Take a real one our engine works through: an LED light-therapy face mask — a plastic shell with red and blue LEDs and silicone padding, sold for consumer skincare. Two defensible-sounding headings compete:
| Candidate heading | Reads as | Column-1 rate |
|---|---|---|
| 8543.70.98 | Electrical apparatus with an individual function | 2.6% |
| 6307.90.98 | Other made-up textile "mask" articles | 7% |
| 9505.90.60 | Festive / costume article | Free |
On a $10,000 entry, the gap between the 2.6% electrical line and the 7% textile line is $440 of base duty on a single shipment — before any surcharge or fee. Call it a "mask" and you may overpay; call it electrical apparatus and you'd better be able to show why. The word "mask" is doing a lot of work here, and it's the wrong kind of work: the device's function is emitting therapeutic light, not covering a face. That's the essential-character argument, and it's the one CBP has actually accepted in its rulings.
Precedent-first: the honest way to pick
Here's the part most tools skip. CBP publishes its classification decisions — tens of thousands of ruling letters in a database called CROSS (rulings.cbp.gov). If CBP has already ruled on a device like yours, that ruling is the answer, or the closest thing to it. So the defensible method is not "ask a model to name a code." It's:
- Search CBP's own rulings for goods that share your product's function and materials.
- Take the candidate codes from those rulings — so every code you consider already carries a real citation. A code with no ruling behind it is a guess wearing a suit.
- Rank candidates by how on-point the rulings are (a ruling about "light therapy" outweighs one that merely shares the word "mask"), and verify each rate live against the current HTSUS.
- Write down which heading you adopted, which you rejected, and why — in GRI terms.
For the LED mask, that method surfaces five CBP rulings placing similar devices under 8543.70.98 (electrical apparatus) and a pile of 6307 rulings that, on reading, cover disposable nonwoven face masks — a different product entirely. The precedent doesn't just suggest a code; it tells you why the 7% textile line is the wrong one for this good.
8543.70.96 when the current live line is 8543.70.98. A dead 8-digit line has to inherit into the current catch-all under the same six-digit subheading — not silently drop its precedent. Any tool that ignores lineage will either miss good rulings or cite dead codes. Ours flags the lineage assumption for human review rather than hiding it.Why "my broker chose it" isn't a defense
The importer of record is legally responsible for the classification — not the broker who filed it. CBP's reasonable-care standard expects you to have consulted the tariff schedule and relevant rulings and to keep a written record. When a CF-28 request-for-information or a CF-29 notice-of-action lands, "my broker picked it" doesn't answer the question CBP is asking, which is: on what basis? A one-line code in an entry summary is not a basis. A written GRI analysis citing the rulings behind the code is.