How NAICS and SIC codes are assigned to companies in commercial datasets

An official NAICS code describes an establishment, a single location, and follows its major activity. SIC codes are still carried in the SEC's records of its filers and in commercial datasets such as Dun & Bradstreet's, which can list several codes for one company with the primary code first. This post shows how each code is assigned, why two datasets code one company differently, and how to test the difference on your own data.

How NAICS and SIC codes are assigned to companies in commercial datasets

In federal statistics a NAICS code is assigned to each establishment, one physical location, by its major activity, and a company's code is derived from its establishments by a stated rule. The SEC records SIC codes for each filer. A commercial dataset codes the company under the vendor's own rule, and may list one primary code with secondary codes.

So a code in a vendor file answers four questions before it names an industry: which system and edition, which unit, who chose it, and whether it stands alone or leads a list. Two datasets that answer any of them differently can disagree about a company with neither in error.

Industry is one field of firmographic data; the comparison of firmographic and technographic data places it beside what a company runs.

The difference between NAICS and SIC industry classification

NAICS replaced SIC for US federal statistics in 1997. It was developed jointly by the United States, Canada and Mexico, groups establishments by how they produce, uses codes of two to six digits and is reviewed for revision every five years. SIC, in its 1987 manual, uses codes of up to four digits and survives in SEC filer records and commercial datasets.

The Office of Management and Budget's 2021 notice on the 2022 edition records the 1997 adoption and calls NAICS production-oriented, grouping units whose production processes are alike. The notice of final decisions names the partners, Mexico's INEGI, Statistics Canada and OMB, and the cycle: revisions are considered every five years, in years ending in 2 and 7.

The official NAICS and SIC sources: Census Bureau, OSHA and the SEC

The Census Bureau publishes NAICS: the 2022 edition, its 20 sectors and the codes beneath them. OSHA hosts the 1987 SIC manual, searchable by keyword or by a two-, three- or four-digit code. The SEC publishes its own SIC list, which names the office that reviews filings in each code.

The Census Bureau's guide to understanding NAICS sets out the levels from sector to national industry, OSHA's SIC search covers the 1987 manual, and its SIC manual lists the ten divisions with their major groups.

The three differ in purpose. The 2021 notice states that NAICS is maintained solely for statistics, and that users who apply it to regulation, administration or tax have no part in its revision. The SEC uses SIC for administration: by its SIC code list, the code shown on a company's filings signals the kind of business it runs, and the Division of Corporation Finance uses that code to decide which office reviews them. The SEC's guide to using EDGAR shows each company's SIC code beside its Central Index Key, and a search by code, such as 7370, returns every registered company filed under it.

The SEC list is its own. It carries 8880, American Depositary Receipts, and 8888, Foreign Governments, while Major Group 88 of the 1987 manual holds a single industry, 8811, Private Households. A field labelled SIC can also be UK SIC 2007, which the Office for National Statistics maintains and which Companies House takes, in a condensed list, when a company files. ONS finalised UK SIC 2026 in April 2026 and names the 2031 Blue Book as its earliest expected use in the national accounts. For NAICS, OMB requested comments on 13 July 2026 on proposed updates for 2027.

Census NAICS code assignment: each establishment by its primary business activity

The Census Bureau assigns a NAICS code to each establishment, a single physical location, and classifies the whole establishment by its major activity. The code comes first from Economic Census and survey answers. Where those are missing, the Bureau draws on administrative records, including BLS classifications and the codes businesses give themselves on income tax returns.

The 2022 Economic Census methodology lists what the Bureau reads on the form: the kind-of-business boxes a respondent ticks, the product lines sold or made, and the respondent's own written description of the establishment. When every critical answer is present, the code is assigned automatically. The County Business Patterns methodology adds a last step: a record classified only in part takes its code from an establishment with similar payroll, employment and type of organisation.

The Bureau of Labor Statistics keeps codes of its own for the establishments in its Quarterly Census of Employment and Wages. Its handbook describes the Annual Refiling Survey, run on a three-year cycle for most establishments, which asks employers about their activity so that each establishment sits in the correct industry, and reclassifies an establishment whose business has changed.

A company with establishments in several industries gets one code only by a rule. For its Statistics of U.S. Businesses, the Census Bureau's glossary sets the rule out. Manufacturers' sales branches, establishments that manage the company (sector 55, Management of Companies and Enterprises) and auxiliary establishments are left out. The enterprise is then placed in the sector where it paid the largest share of its payroll, then by the same test in a subsector and a four-digit industry group. Payroll decides the company's code there, and the result stops at four digits.

Why two datasets give one company different codes

Four causes of disagreement can be read from the documentation of the two sources, and none of them is an error by either.

• Company against establishment. The Census Bureau codes each location and, for its business statistics, rolls a company up by payroll. A vendor codes the company record. A factory and a software office in different industries give two honest answers, and the corporate hierarchy adds more, since a subsidiary can be coded apart from its parent.
• Self-reported against assigned. A code a business gives itself, on a tax return or a Companies House filing, is its own choice. The Census Bureau uses such self-assigned tax codes when its own survey answers are missing. A vendor's code may come from the business, from a public record, from its analysts or from a model.
• One code against several. A statistical roll-up returns one code. A vendor may keep a primary code, chosen by revenue, and secondary codes. Comparing one source's single code with the other's second code finds agreement that a comparison of primaries misses.
• The edition of the system. The 2022 edition moved Software Publishers from 511210 to 513210, so two sources a revision apart can show different codes for one activity. Editions also lag: the County Business Patterns methodology lists its 2017 to 2023 data as using 2017 NAICS. And a SIC field may hold the 1987 manual, the SEC's list, UK SIC 2007 or a vendor's extension.

How Dun & Bradstreet assigns SIC and NAICS codes

Dun & Bradstreet's documentation describes a company record that can carry several industry codes. In its profiles and credit reports, the first SIC code listed is the primary line of business, and the codes after it account for at least 10% of the firm's annual revenue. Its API documentation gives each code a sales percentage.

Its guide to SIC and NAICS codes says it provides both systems in its products, such as D&B Hoovers company profiles and D&B Finance Analytics credit reports. Its Direct 2.0 documentation for company data describes an industry code entry that can repeat within one record, with the code in first position as the primary one, reflecting the largest share of revenue.

Its documentation for finding industry codes says the codes are generally set by an external authority, such as the US Office of Management and Budget, and that Dun & Bradstreet may extend them to define an activity more precisely: the last four digits of its 8-digit SIC, which it calls SIC 2+2, are such an extension. The code types it lists include NAICS, UK SIC 2007 and its own D&B Standard Industry Code. For a join to another source, compare primary with primary, and cut an 8-digit SIC to its first four digits before matching it to the SEC's list or the 1987 manual.

Example: one company, two datasets, two primary codes

Acme Robotics is an illustrative US company that builds robots for general industrial use and publishes the software that runs them. Two datasets code it, and they disagree. Assumptions:

• Acme Robotics has three establishments: a factory, a software office and a head office.
• Annual revenue: factory $60 million, software office $40 million, head office none.
• Annual payroll: factory $14 million, software office $18 million, head office $8 million.
• The software office publishes software sold as a product: Software Publishers in NAICS (511210 in the 2017 edition, 513210 in 2022) and Prepackaged Software in SIC (7372).
• Source A codes a company with the NAICS 2022 code of its establishment with the largest payroll, leaving head offices out as the Census Bureau's enterprise rule does.
• Source B codes a company in SIC: the activity with the largest share of revenue first, then every other activity with at least 10% of revenue.

The working:

1. Source A. Without the head office, payroll is $14 million plus $18 million, $32 million. The software office pays $18 million of it, 56.25%, so Acme Robotics is coded as a software publisher: 513210 in NAICS 2022, in sector 51, Information. With the head office counted, the software office's share is $18 million of $40 million, 45%, against 35% for the factory, so the answer holds.
2. Source B. Revenue is $60 million plus $40 million, $100 million. Robots bring 60%, so the primary code is 3569, General Industrial Machinery and Equipment, NEC, the industry under which the 1987 manual lists robots for general industrial use. Software brings 40%, above 10%, so 7372, Prepackaged Software, follows as a secondary code.
3. The comparison. The primary codes name different activities. Source A's activity, the software office, appears in Source B's list as its secondary code. The example assumes 513210 and 7372 describe that one activity; on real data, check such a pairing against the official concordance before you rely on it.

Both sources are right by their own rules. Had the software office paid $13 million rather than $18 million, Source A would have coded the factory and the two would agree on the primary activity. A third source still on NAICS 2017 would show 511210 where Source A shows 513210: the same industry, numbered as before the 2022 edition.

A test for comparing industry codes across two sources

Run this test after entity resolution has matched the records of the two sources to one company. It turns a raw match rate into agreement you can explain, cause by cause.

1. Label each code with its system and edition: NAICS 2017, NAICS 2022, the 1987 SIC manual, the SEC's list, UK SIC 2007 or a vendor's extension. Map the two sides to one edition through an official concordance before comparing.
2. Label the unit: an establishment, an enterprise rolled up from establishments, an SEC filer or a vendor's company record. Roll establishments up by a stated rule, such as the Census Bureau's payroll rule, before comparing them with a company code.
3. Label who chose the code: the business itself, a statistical agency, the vendor or a model.
4. Compare primary with primary, then each primary with the other side's full list. Score full agreement, partial agreement when one primary appears among the other side's secondary codes, and disagreement.
5. Where both sides use NAICS, record the deepest level at which they agree: six digits, then five, four, three and the two-digit sector. Read the ranges 31-33, 44-45 and 48-49 as one sector each.
6. Take 20 disagreements, read each company's own description of its business, and record which of the four causes explains each one.

Two sources that disagree on primaries but agree on full lists hold the same activities and rank them differently. That finding says how to join the sources, not which one to drop.

Industry in Fokals data

In the dashboard and the data browser, each company website carries an industry label, one of 21 industries. An empty value reads as undecided.

The same industry is the industry dimension of Market Series, the weekly series across 21 industries, countries, company size bands and markets, which also cut by sector, carried as the company's sector text. Each private capital raise in Company Funding, part of the intent data, carries the industry group the issuer gave in its regulatory filing, with the issuer's SEC CIK. A listed company in Fokals data carries its SEC CIK where it has one, and the SEC's company search returns the SIC code it records for that filer, so you can hold both classifications side by side.

Under the test above, record the website label as a system of its own, set for each website and versioned, in its own column beside any NAICS or SIC code you hold. Every field named here is defined in the data dictionary.

Frequently asked questions

What this page says about the products it names was checked against their public documentation on 7 October 2026. Product and company names are trademarks of their owners. Fokals is not affiliated with them or endorsed by them.