How many restaurants are in NYC? A 2026 borough count

The rounded figure comes from NYC Health's restaurant inspection page, checked October 7. The exact count below is Marji's tabulation of the official DOHMH inspection dataset. We counted each establishment identifier once, then separated records with an inspection from records using the city's not-yet-inspected placeholder. The distinction matters if you are sizing a neighborhood, writing a business plan or citing a citywide total.

NYC restaurant records by borough

With inspection means the record has at least one inspection date later than January 1, 1900. Not yet inspected means its latest inspection date is January 1, 1900, the placeholder DOHMH assigns to new establishments awaiting a first inspection. The borough-unassigned row preserves source code 0; moving those 95 records into a borough would require information we have not verified. All three columns reconcile to the source's distinct-establishment count. That live link will change as the city updates the dataset.

Why 29,000 and 31,469 can both appear in an answer

The city's public introduction supplies a rounded restaurant count. The downloadable inspection dataset supplies a specific administrative population on a specific date. Its published scope includes restaurants and college cafeterias in active status, plus applicants that have not received an inspection. It retains inspection and violation history for those establishments. Adding all those rows would count the same business repeatedly.

We cannot reconcile the rounded city figure to this exact extract without a matching date and definition from NYC Health. Nor should subtracting the 3,858 uninspected records be presented as the definitive answer: the remaining 27,611 are establishments with recorded inspections, not 27,611 confirmed open restaurants. A precise number is useful only when its label stays attached.

For the employment question, BLS's Quarterly Census of Employment and Wages counts establishments, jobs and wages within its unemployment-insurance reporting coverage. That is a different unit and reporting system. A quarterly employer-establishment count should not be added to, averaged with or treated as a correction to this health-inspection count.

What the borough totals tell an operator

Manhattan accounts for 39.9% of all records in this extract: 12,565 divided by 31,469. Brooklyn accounts for 25.4%, and Queens 22.4%. Those are shares of administrative records. They do not measure shares of restaurant revenue, demand, seats or independent ownership, and this analysis does not classify chains separately.

For an operator choosing a site, use the borough table to frame the search, then build a much smaller comparison set. A lunchtime counter competes for a different visit from a reservation-led dining room. Define a walkable area or delivery radius, inspect current menus and opening hours, and check the actual sites. Record the dates: an old inspection can be useful history while saying little about today's menu or service.

• —Match the occasion. Compare restaurants serving the meal, price point and customer visit you plan to serve. A borough-wide count cannot do that filtering for you.
• —Verify the location. Check the address and whether it is operating. An identifier is an administrative unit; a list of identifiers is not a field survey of storefronts.
• —Build your own economics. Estimate rent, staffing, ingredient cost and required sales. The break-even guide turns those assumptions into a sales threshold; the New York wage guide covers the dated wage inputs.
• —Keep the menu comparable. Note portion, channel and included sides before comparing prices. Posted prices do not reveal a competitor's supplier costs or profit.

Can this dataset show NYC restaurant closures?

This snapshot cannot establish a permanent-closure total or a restaurant failure rate. The dataset description says its population is establishments in active status on the record date. Historical inspections within that population do not reconstruct every restaurant that existed in an earlier year.

An inspection action recording a health-department closure also answers a different question from a permanent business shutdown. A later reopening may appear in the same establishment's history. Counting closure-action rows would mix repeat actions and businesses with different subsequent outcomes. For the meaning of grades and the limits of the inspection record, use our NYC restaurant inspection grades analysis alongside the city's current lookup.

Measuring openings and closures over time would require comparable saved snapshots, consistent identifiers and a way to distinguish an administrative status change from a real business event. Without that work, a claim such as 'NYC lost 500 restaurants' is not supported by comparing today's total with an older number found elsewhere.

Method: how we counted, and how to reproduce it

1. 1.Select the official source. Use DOHMH dataset 43nn-pn8j, not a community-created filtered view. Our October 7 pull contained 296,270 rows with record timestamps between 06:00:15 and 06:00:21 on October 6, 2026. The source supplies those timestamps without a timezone.
2. 2.Count establishments once. Group by camis and boro, then take max(inspection_date). CAMIS is DOHMH's unique establishment identifier. Check the result against an independent count(distinct camis) query. Every identifier in this extract appeared in only one borough group.
3. 3.Separate inspection status. A maximum inspection date of 1900-01-01 goes in the not-yet-inspected column. A later date goes in the with-inspection column. We found no missing dates in these grouped results; the most recent actual inspection date was October 4.
4. 4.Preserve unknown geography. Keep borough code 0 as an unassigned row. Add the six rows to obtain each citywide column total.
5. 5.Check extraction consistency. The source update timestamp was unchanged before and after the pull. The 31,469 grouped records matched the distinct-identifier query. Percentages use the stated denominator and round to one decimal place.

The Socrata API documentation provides the query interface. The essential selection is camis, boro, max(inspection_date), grouped by camis, boro; request enough rows to include every group. These published numbers are a fixed dated observation. Re-running the query later should produce a fresh observation, not silently overwrite the period behind this table.