How the Bureau of Labor Statistics builds the monthly jobs report

What to take away

• The latest news jobs number is a survey estimate, not a headcount, and it comes from two separate BLS surveys that often disagree.
• The establishment survey (Current Employment Statistics) produces the nonfarm payrolls figure from a sample of employer payroll records.
• The CPS household survey produces the unemployment rate from a sample of about 60,000 households.
• Both surveys cover a reference period that includes the 12th of the month, and the first print arrives about two weeks later.
• The first payrolls figure gets revised twice, once the following month and once the month after that.
• Seasonal adjustment and the birth-death model remove recurring patterns and account for businesses that open or close between sample updates.

What the monthly jobs report actually measures

The Employment Situation release, usually called the jobs report, answers several questions at once. How many people worked for pay during a specific pay period? How many were actively looking for work and could not find it? How many hours did they work, and what did they earn?

The Bureau of Labor Statistics, the federal statistical agency that produces the report, draws those answers from two surveys with different designs, samples and definitions. One surveys employers. The other surveys households. They are not two views of the same dataset. They are two datasets.

The headline number most readers see is nonfarm payrolls, the month over month change in jobs on employer payrolls. The second headline is the unemployment rate, the share of the labor force that is jobless and actively looking. Both appear in the same release, and reporters often quote them in the same sentence. They come from different instruments.

The report also includes average hourly earnings, the average workweek, the labor force participation rate, the employment population ratio, and counts of people unemployed for different lengths of time. Each has its own sample and its own quirks.

One consequence matters for anyone reading the latest news: the two headline numbers can move in opposite directions in the same month. Payrolls can rise while the unemployment rate also rises, because the two surveys count different things.

Inside the BLS establishment survey and its sample of employers

The establishment survey, formally the Current Employment Statistics program, samples employer payroll records rather than asking people about their own work. The Current Employment Statistics - CES (National) : U.S. Bureau of Labor Statistics page describes the program as a monthly survey of about 121,000 businesses and government agencies representing roughly 631,000 worksites.

Those employers report the number of people on their payrolls during the pay period that includes the 12th of the month, along with paid hours and payroll earnings. BLS collects the data by mail, phone and secure electronic reporting, and the sample rotates so no employer carries the same burden forever.

The payroll count excludes several groups by design. It leaves out agricultural workers, private household employees, unpaid family workers, proprietors, and most of the self employed. It counts jobs, not people, so a worker holding two jobs appears twice.

That last point explains why payrolls can exceed the number of employed people in the household survey. It also explains why a single person picking up a second job raises payrolls without changing the count of employed workers.

The establishment survey supplies the nonfarm payrolls headline, the average hourly earnings figure and the average weekly hours figure. It is the source of the number that moves markets within seconds of the 8:30 a.m. Eastern release.

The household survey and why it can diverge from payrolls

The household survey is the Current Population Survey, a joint project of BLS and the U.S. Census Bureau. The CPS Home : U.S. Bureau of Labor Statistics page explains that it interviews about 60,000 households each month and classifies each person 16 and older as employed, unemployed or not in the labor force.

A person counts as employed if they did any work for pay or profit during the survey week, or worked at least 15 hours as an unpaid family worker. A person counts as unemployed if they had no job, were available to work, and actively looked for work in the prior four weeks.

Anyone else is not in the labor force.

The Frequently Asked Questions (FAQs) : U.S. Bureau of Labor Statistics page spells out those definitions and how they differ from common usage.

The unemployment rate is unemployed people divided by the labor force, which is the employed plus the unemployed. People who have stopped looking are outside the labor force and outside the rate. That is why the rate can fall because people found work, or because they gave up looking.

The household survey counts people, not jobs, and it includes the self employed, agricultural workers and private household workers that payrolls exclude. It also asks about multiple jobholders, part time work for economic reasons, and how long people have been out of work.

Because the two surveys use different samples, different questions and different definitions, they can diverge for months at a time. When they do, the divergence is usually explained by those design differences rather than by an error in either survey.

For a wider comparison of how this release sits alongside price and output data, see our economic indicators compared piece.

The reference period and the collection window behind the headline number

Every monthly jobs report describes a specific slice of time. The establishment survey uses the pay period that includes the 12th of the month. The household survey uses the calendar week that includes the 12th, typically Sunday through Saturday.

That reference period matters because it fixes the moment the data describe. A strike that starts on the 20th, a hurricane that makes landfall on the 25th, or a hiring freeze that begins late in the month may not appear in that month's figures at all. It shows up in the next release.

The collection window runs after the reference period. Employers report payroll data over the following weeks, and Census Bureau interviewers conduct household interviews during the week after the reference week, with follow up calls for households that did not respond.

BLS then edits, weights and tabulates the responses. The Employment Situation release lands about two weeks after the reference period closes, usually on the first Friday of the following month. The Employment Situation Summary - 2026 M08 Results shows the standard structure: payrolls, the unemployment rate, earnings, hours, and dozens of tables.

Reporters who cover the release on deadline often work from the summary and the first few tables. Knowing the reference period tells you which events could plausibly be in the number and which could not. Our jobs report on release day desk case walks through that judgment call.

Seasonal adjustment and the birth-death model in plain terms

Hiring follows the calendar. Retailers add workers before the winter holidays. Construction picks up in spring. School districts hire in late summer and shed staff in June. Those patterns repeat closely enough that raw, unadjusted numbers would mislead anyone comparing one month with the next.

Seasonal adjustment removes the part of the monthly change that can be predicted from the calendar. BLS estimates seasonal factors from years of history and applies them to the current data. The result is the seasonally adjusted figure that dominates coverage.

Unadjusted figures still appear in the tables, and they are the right choice for comparing the same month across years.

Seasonal adjustment is not a judgment about whether the economy is good. It is a filter. The filter itself changes as new data arrive, and BLS revises seasonal factors once a year, which can shift recent seasonally adjusted figures.

The birth-death model is a separate adjustment. The establishment survey sample cannot capture businesses that opened or closed since the sample was last updated, so BLS models the net job creation from those births and deaths using historical patterns.

The model is not a guess about any particular startup. It is an estimate of the net contribution of firms outside the sample.

That estimate can be wrong in turning points. In a recession, business deaths may exceed the model's expectation, and the model may temporarily overstate payroll growth. BLS publishes the model's contribution so analysts can see how much of a month's change comes from it.

Definitions and methods for all of these adjustments appear in the agency's own materials, and our BLS glossary covers the terms reporters trip over most.

Why the first jobs print gets revised twice

The first payrolls figure is preliminary. It comes from the employers that reported in time for the initial tabulation, weighted to represent the full population. Late reports, corrections and newly added sample units arrive afterward, and BLS folds them in.

The first revision comes with the next month's release. The second comes with the release after that. By then the estimate has absorbed most late responses, and the number is considered final until the annual benchmarking process, which re-anchors the level of payrolls to near complete unemployment insurance tax records.

Here is the sequence in plain steps:

1. BLS publishes the advance estimate for month one, based on responses received by the collection cutoff.
2. One month later, BLS publishes the first revision to month one alongside the advance estimate for month two.
3. One month after that, BLS publishes the second revision to month one alongside the advance estimate for month three.
4. Once a year, BLS benchmarks the payroll series to unemployment insurance tax records and may revise several years of data.

The size of those revisions varies. Most months the first two revisions are small relative to the headline change, but in some months they are large enough to change the story. A month reported as a gain can become a smaller gain or, less often, a decline.

The unemployment rate from the household survey is also revised, but its revisions are usually smaller and come from population controls and seasonal factor updates rather than from late employer reports. The two headline numbers follow different revision paths.

Feature	Establishment survey (CES)	Household survey (CPS)
Who is asked	About 121,000 employers, roughly 631,000 worksites	About 60,000 households
Unit counted	Jobs	People
Reference period	Pay period including the 12th	Week including the 12th
Headline produced	Nonfarm payrolls, earnings, hours	Unemployment rate, participation
Includes self employed	No	Yes
First revision	Next month's release	Usually small, from controls
Second revision	Release after that	Annual seasonal updates

How to read the latest news jobs report without overreading one month

One month of payrolls is a noisy signal. The monthly change routinely swings by tens of thousands of jobs from one report to the next without any change in the underlying trend. Reporters who treat a single print as a verdict on the economy will be wrong often.

The standard fix is to average. Three month and twelve month averages smooth the month to month noise and show the direction of hiring. BLS publishes both in the release, and they are usually the better number to quote.

Watch the revisions as closely as the headline. If the prior two months were revised down by a combined 100,000, a strong current month may not mean what it appears to mean. The release shows those revisions in the first table.

Check the household survey details. The unemployment rate can hold steady while the labor force shrinks, or fall while employment falls. Participation and the employment population ratio tell you which is happening.

Read the establishment survey details too. Average hourly earnings can jump because of composition, such as a month with fewer low wage retail jobs, rather than because employers raised pay. Average weekly hours can signal weakening demand before layoffs appear.

Use the definitions consistently. The About Us : U.S. Bureau of Labor Statistics page describes the agency's role as an independent statistical producer, which matters when a release contradicts a political narrative. The numbers are estimates with stated margins of error, not verdicts.

Before publishing, run the release through a disciplined process. Our economic release checklist covers the steps for the morning a number lands, and our economic reporting problems piece covers the mistakes that show up most often in coverage.

• Confirm which reference period the report covers.
• Quote the seasonally adjusted headline and note the unadjusted figure if relevant.
• Report the prior two months' revisions.
• Use the three month average for trend, not a single month.
• Check participation and the employment population ratio alongside the unemployment rate.
• Attribute all figures to the specific BLS release and table.
• Note the margin of error or the survey's estimated variance where the story turns on a small change.

Common questions

What is the difference between the establishment survey and the household survey? The establishment survey asks employers about jobs on their payrolls, while the household survey asks people about their own work and job search. They use different samples, definitions and reference periods, so they can produce different pictures of the same month.

Why does the unemployment rate sometimes rise while payrolls rise? The two numbers come from different surveys. Payrolls count jobs reported by employers, and the unemployment rate depends on how many people are in the labor force and actively looking. People entering the labor force to search can raise the rate even as hiring continues.

How large are the revisions to the first payrolls print? Most revisions are modest, but they can be large enough to change the direction of the reported change. The first revision arrives with the next month's release, and the second arrives the month after that.

What is the birth-death model and why does it exist? It estimates the net jobs added by businesses that opened or closed since the sample was last updated. Without it, the payroll count would miss firms that are too new to be in the sample.

Does seasonal adjustment change the underlying data? It changes the presentation, not the raw responses. BLS publishes both seasonally adjusted and unadjusted figures, and the unadjusted series is the right one for comparing the same month across years.

Where can I find the definitions BLS uses? The agency publishes its concepts and definitions in the CPS documentation and in its public FAQ pages, which cover who counts as employed, unemployed or outside the labor force.