Nice Classification for Software — Class 9, Class 42, and the Class 35 Mistake.

What the Nice Classification is.

The Nice Classification is an international system administered by WIPO that groups goods and services into 45 classes — classes 1–34 for goods, classes 35–45 for services. The system is named after the 1957 Nice Agreement, which established it. The current edition is updated periodically; the relevant edition for an application is the one in force at filing.

The USPTO maintains its own Acceptable Identification of Goods and Services Manual — the ID Manual — that lists pre-approved identifications for each class. Filing with ID Manual language is the fast path through examination; filing with custom language triggers an examiner’s identification refusal that costs weeks. For most software identifications, the ID Manual contains a phrase that fits the goods, and using the manual’s language is the right move.

Class 9 — Downloadable software, recorded media.

Class 9 is the canonical class for goods that ship as software objects: downloadable applications, mobile apps, software embedded in hardware, recorded media (music, video), and a long tail of related technical apparatus. For a product whose customer downloads a binary — an iOS app, an Android app, an Electron desktop app, an installable executable — Class 9 is the class.

The pre-approved ID Manual phrasing for downloadable software typically reads: “Downloadable software for [specific purpose] in the field of [specific industry].” The two bracketed slots are where applicants tend to over-claim. The temptation is to write “Downloadable software for managing data” — broad, defensible, future-proof. The cost of that breadth is that the examiner reads it against every senior mark with similar breadth, and the §2(d) collision risk expands proportionally. The discipline is to specify enough that the examiner can fence your goods from the senior marks’ goods.

Class 9 is also the densest class for software disputes. Of the trademark applications filed by venture-backed startups in our outbound corpus over the last 14 months, 73% included Class 9 and 41% drew at least one §2(d) refusal tied to that class. The class is crowded and litigious.

Class 42 — SaaS, technological consultancy.

Class 42 is the canonical class for cloud-hosted services: software-as-a-service, platform-as-a-service, technology consulting, data-analytics services, and scientific and engineering services. For a product whose customer accesses functionality via a hosted web service — with no binary download — Class 42 is the class.

The pre-approved ID Manual phrasing for SaaS typically reads: “Software-as-a-service (SAAS) services featuring software for [purpose] in the field of [industry].” Same discipline as Class 9: specify enough to fence your goods.

Most modern AI and SaaS startups need both Class 9 and Class 42 — Class 42 for the hosted product, Class 9 for the downloadable client (mobile app, desktop app, CLI tool). Filing in both classes is fully standard. The USPTO filing fee is per class, currently $350 per class for the TEAS Standard option, so a two-class filing costs $700 in filing fees plus prosecution. The cost is small relative to the protection breadth.

The Class 35 mistake.

Class 35 covers advertising, business management, business consulting, retail and wholesale store services, marketing, and office functions. It is the wrong class for software, and it is the class most founders mistakenly file in. The mistake happens because the word “business” appears in the class title and Class 35 sounds like a natural home for business software. It is not.

A business-software product shipped as downloadable code belongs in Class 9. The same product shipped as a hosted service belongs in Class 42. Class 35 covers the services of running someone else’s business — advertising agencies, management consulting firms, business-process outsourcing — not software that helps a business run itself.

If you file in Class 35 by mistake, the examiner will typically issue an identification refusal asking you to reclassify into Class 9 or 42. Reclassification is a paperwork cost (typically $100–$300 in legal time) but doesn’t kill the application. The deeper cost is that during the months between filing and the reclassification, your Class 35 record returns clean for any Class 9 clearance searches a competitor runs — and any senior marks in Class 9 you were ignorant of remain ignorant of you.

Adjacent classes that sometimes matter.

Beyond Class 9 and 42, a small number of other classes are occasionally relevant for software companies.

• Class 41 — education, training, entertainment. Relevant for edtech, coding bootcamps, content businesses, gaming companies. A game streamed on a Class 42 platform is still Class 41 entertainment.
• Class 45 — legal services, security services. Relevant for legal-tech and security companies, when the goods include human-provided services on top of the software layer.
• Class 36 — financial services, insurance, banking, real-estate services. Relevant for fintech companies whose product includes regulated financial services in addition to software.
• Class 44 — medical services, agricultural services, horticultural services. Relevant for healthtech companies whose product includes medical-services delivery in addition to software.
• Class 39 — transportation and logistics services. Relevant for logistics software companies whose product includes actual transportation services.

The pattern: if your business mixes software with a regulated or non-software service, you likely need filings in multiple classes covering each surface. A healthtech company that sells a SaaS platform and delivers telemedicine consultations would typically file in Classes 9, 42, and 44. Companies whose product brushes against famous-mark territory in any of these classes should also read Famous Mark Rules — the famous-mark exception cuts across class lines.

The ID-of-goods discipline.

Within a class, the ID-of-goods language defines what your registration actually covers. Two applications in Class 42 with different ID language give different protection. Five principles guide good ID drafting.

1. Use the USPTO ID Manual phrasing where possible. Pre-approved language sails through examination. Custom language triggers an identification refusal.
2. Specify the purpose, not the technology. “Software for managing customer relationships” is good. “Software using natural-language processing and machine learning” is bad — the technology stack will change before your registration matures.
3. Specify the field, not the buyer demographic. “In the field of construction project management” is good. “For users aged 25–45” is bad and likely rejected.
4. Be narrow enough to differentiate, broad enough to cover the roadmap. The discipline is to specify enough to fence away the senior marks but not so much that a competitor pivot to an adjacent feature falls outside your protection.
5. Match your actual use. If you claim use in commerce (“use-based” application, basis 1(a)) the goods must match what you’re actually doing as of the filing date. If you file on intent-to-use (basis 1(b)) the goods must match what you’ll be doing when you file the Statement of Use.

When the Nice class travels internationally.

The Nice Classification is harmonized internationally — the Class 9 and Class 42 you file in the U.S. are the same Class 9 and Class 42 you’d file in the EU, UK, Canada, Australia, Japan, or any other Madrid Protocol member. The ID-of-goods language often is not harmonized — different offices have different ID conventions and different views of what an acceptable identification looks like.

For founders planning international filings via the Madrid Protocol, the practical consequence is that the basic U.S. application’s class structure transfers cleanly but the ID language may need to be tightened or loosened for individual designated countries. We cover the Madrid mechanics at Madrid Protocol — One Application, 130 Countries.

How Etymolt verifies trademarks.

Etymolt’s trademark axis maps every candidate against the full Nice Classification, with the heuristics tuned for software companies. The default clearance run hits Classes 9 and 42 with explicit ID-of-goods language analysis; the engine also surfaces neighboring-class collisions where the goods-relatedness language suggests cross-class confusion risk. Every flag carries the underlying USPTO record number. The verdict is attested and permalinked. Full methodology at /methodology. If you’re new to clearance, start with How to Check If a Brand Name Is Trademarked.