The jobs report: how to read nonfarm payrolls and the two surveys

The jobs report is the US Bureau of Labor Statistics' monthly Employment Situation release. It is the first broad, official reading of the labor market for the month just ended, and it usually lands on a Friday early in the following month at 8:30 a.m. Eastern. The October 2026 report, for example, is scheduled for Friday, November 6, 2026. Its headline number, the monthly change in nonfarm payrolls, is probably the most watched single statistic in US economics.

The report matters to investors because jobs drive income, income drives spending, and the Federal Reserve's mandate names maximum employment alongside stable prices. A strong or weak report changes what traders expect the Fed to do, and that moves Treasury yields, the dollar and stock-index futures within seconds of 8:30. Yet the headline is also one of the noisiest numbers in finance: it comes from a sample, it is revised twice in the following months and again once a year, and its error band is wider than many of the monthly changes it reports.

This guide explains where the numbers come from, why the payroll count and the unemployment rate can disagree, how revisions and the birth-death model work, and how to read a release without overreacting to it. It uses the BLS data as published through September 2026.

What the jobs report contains

The Employment Situation is a long document: a summary, dozens of tables and a technical note. Most readers use five numbers from it:

• Nonfarm payroll employment: the number of jobs on employer payrolls, outside farming, and its change from the previous month. September 2026: 159.0 million jobs, +29,000 on the month (first estimate).
• The unemployment rate: the share of the labor force without a job and looking for one. September 2026: 4.2%. The The unemployment rate: U-3, U-6 and labor force participation guide covers the definitions and the broader U-1 to U-6 measures.
• Average hourly earnings for private-sector employees: $37.81 in September 2026, 3.0% above a year earlier. The Wage growth: average hourly earnings, real wages and composition guide explains why this series can mislead.
• The labor force participation rate: the share of the adult population working or looking for work. September 2026: 61.8%.
• Revisions to the two previous months' payroll numbers, which can change the story as much as the new month.

Beneath those headlines sit payrolls by industry, average weekly hours, the number of people working part time who want full-time work, long-term unemployment, and unemployment by age, sex, race and education. For a quick definition of the headline, see Nonfarm payrolls.

Two surveys, two different questions

The most important thing to know about the jobs report is that its two headline numbers come from two separate surveys, run differently and asking different questions.

The establishment survey (payrolls)

The Current Employment Statistics (CES) survey asks employers how many people were on their payroll in the pay period that includes the 12th of the month. Each month BLS collects data from about 119,000 businesses and government agencies covering about 622,000 worksites, which together account for roughly a quarter of all nonfarm payroll jobs. Because the sample is so large, the payroll estimate is the more precise of the two headline measures, and it also supplies hours worked and hourly earnings by industry.

It counts jobs, not people. A person with two part-time jobs at two employers appears twice. It excludes farm workers, the self-employed, private household workers and unpaid family workers, and it counts nobody on unpaid leave.

The household survey (unemployment)

The Current Population Survey (CPS), conducted for BLS by the Census Bureau, asks about 60,000 households what each member aged 16 and over was doing during the calendar week that includes the 12th. From the answers, BLS classifies each person as employed, unemployed or not in the labor force:

• Employed means doing any paid work that week, working in one's own business or farm, doing at least 15 hours of unpaid work in a family business, or being temporarily absent from a job.
• Unemployed means having no job, being available for work, and having actively looked for work in the previous four weeks. People waiting to be recalled from a layoff count even if they are not searching. Whether someone receives unemployment benefits plays no part.
• Everyone else, including retirees, full-time students and people who have stopped looking, is not in the labor force.

The household survey counts people, not jobs, so a multiple jobholder is counted once, and it includes the self-employed and farm workers that the payroll survey leaves out. Its sample is far smaller, so its month-to-month changes are much noisier.

Why the two can disagree

Because they measure different things from different samples, the two surveys often point in different directions for a month or two. Payrolls can rise while household employment falls, for instance when people take second jobs (more jobs, same number of workers) or when self-employment drops. Over long periods the two broadly agree; month to month, treat the payroll change as the better measure of job growth and the household survey as the source of the unemployment rate and of who is working.

How the payroll number is estimated

Seasonal adjustment

Employment swings every year for reasons that have nothing to do with the economy: retail hiring before the holidays, school staff leaving in June, construction slowing in winter. The headline payroll and unemployment figures are seasonally adjusted to remove those regular patterns, so that a December change reflects more than the usual holiday hiring. BLS revises the seasonal factors for the last five years once a year, which is one reason older months shift slightly every annual update. The Seasonal adjustment and data revisions: why the numbers change guide covers the method.

The birth-death model

A sample of existing employers has a blind spot: a brand-new business can't be in the sample yet, and a business that closes simply stops reporting. BLS handles this in two steps. First, when a sampled business stops reporting because it went out of business, it is treated as having the same employment trend as the rest of the sample rather than as a loss of every job. That roughly offsets the jobs that new, unsampled businesses create. Second, an ARIMA time-series model, built on the past five years of unemployment-insurance records plus current sample information, estimates the net residual that the first step misses.

The birth-death adjustment is reasonable most of the time, because business openings and closings follow steady patterns. It is least reliable at turning points, when the economy shifts faster than a model trained on the past five years can follow. That is one reason the payroll figures are revised so much around recessions.

Revisions: three estimates and a benchmark

Every payroll number is published three times in consecutive months. The first estimate is based on the responses collected by the deadline; the second and third add late responses. BLS considers a month final only after those two revisions.

Then, once a year, the whole series is benchmarked. BLS compares its March estimate with a near-census count of jobs from the unemployment-insurance tax records that almost every employer files, and it adjusts the series to match. Over the ten years before the current technical note was written, the average absolute benchmark revision to total nonfarm employment was 0.2%, in a range of −0.4% to +0.3%. On a base of about 159 million jobs, 0.2% is roughly 320,000 jobs, nearly five months of payroll growth at the pace seen in 2026. BLS publishes a preliminary estimate of the coming benchmark in late summer; in 2026 that release was scheduled for August 28.

The recent record

The bar chart shows the 24 monthly changes from October 2024 to September 2026, as the data stood on October 30, 2026. Three things stand out.

First, the pace slowed sharply. Average monthly payroll growth, from the same data, was:

For comparison, the average in every year from 2011 through 2019 fell between +165,000 and +249,000.

Second, negative months became common. Seven of the 24 months were declines: January, June, August, October and December 2025, and February and July 2026. The largest was −156,000 in February 2026, followed by −140,000 in October 2025.

Third, the monthly numbers swing widely around the trend. February 2026 (−156,000) was followed by March (+214,000). A swing like that sits well within what sampling error, weather, strikes and seasonal adjustment can produce, which is why analysts look at three-month averages. The three months to September 2026 (−10,000, +133,000, +29,000) average +51,000.

The long view

The level chart puts the monthly changes in context. Payrolls rose from 140.6 million in January 2015 to 152.3 million in February 2020. They then fell to 130.4 million in April 2020, a loss of 21.9 million jobs in two months; April alone shows −20.5 million in the current data. Employment regained its February 2020 level in June 2022 (152.4 million), about two years and four months later.

That recovery was far faster than the one after the 2007–2009 recession. Payrolls peaked at 138.4 million in December 2007, fell to 129.7 million by February 2010 (−8.7 million), and did not regain the 2007 peak until May 2014, more than six years after it.

Since mid-2022 the line has kept rising, but its slope flattened in 2025. At 159.0 million in September 2026, payrolls were about 6.8 million above their pre-pandemic peak.

Statistical noise: how big a change is real?

Because the payroll figure comes from a sample, every monthly change has a margin of error. BLS states that the 90% confidence interval on the monthly change in total nonfarm employment is about ±122,000. Its own example: a reported gain of 50,000 means the true change probably lies between −72,000 and +172,000, a range that includes zero, so the gain can't be called statistically significant. A gain of 250,000 would be.

The household survey is noisier still. At an unemployment rate near 6%, the change in the number of unemployed needs to be about 425,000 to be significant, and the change in the rate about 0.3 percentage point.

Error bands are why markets and economists lean on other labor data to confirm a trend: weekly Jobless claims: reading the weekly unemployment insurance report, the openings and quits in JOLTS: job openings, quits and what labor turnover reveals, and the unemployment trend rule described in The Sahm rule: a recession signal from the unemployment rate.

How to read a jobs report

Start with the surprise, not the level

Before each release, economists surveyed by news organizations publish a consensus forecast for payrolls and the unemployment rate. Prices already reflect that consensus, so the market reacts to the difference between the actual number and the forecast. A +50,000 print can be "strong" if the consensus was −20,000 and "weak" if it was +150,000. The How markets react to economic data: surprises, regimes, revisions guide covers this in detail.

Add up the revisions

Net revisions to the previous two months are part of the news. A new month that beats forecasts by 40,000 alongside a combined downward revision of 60,000 to the two prior months leaves the labor market weaker than traders thought the day before.

Check the pieces against each other

A useful order of reading:

1. Payroll change and the three-month average. The trend matters more than any one month.
2. Revisions to the two prior months.
3. The unemployment rate and why it moved. It can fall because people found jobs, or because people stopped looking and left the labor force. Participation tells you which.
4. Average hourly earnings. Wage growth feeds into inflation and Fed thinking, but shifts in who is employed can distort it.
5. Breadth. Is hiring spread across industries, or concentrated in one or two sectors?
6. Hours. Employers often trim hours before they cut jobs, so a falling average workweek can be an early warning.

Translate it into Fed expectations

For markets, the jobs report is mainly an input into the Fed's next move. A soft report can raise the odds of rate cuts, or reduce the odds of hikes; a hot one does the opposite. Rate-sensitive assets therefore move first: the 2-year Treasury yield, which tracks expected Fed policy closely, and the dollar. The effect on stocks depends on which force dominates: lower expected rates help valuations, but weaker jobs also mean weaker earnings ahead. The How Fed decisions move markets: expectations, surprises and yields guide follows that chain.

Pitfalls and limitations

• Treating the first print as fact. The July 2024 example shows that a first estimate can be cut in half by later revisions. The direction of revisions also tends to persist around turning points, when the birth-death model and late responses lag reality.
• Ignoring the error band. Monthly changes inside ±122,000 are not statistically distinguishable from zero on their own.
• Reading the unemployment rate without participation. A falling rate can reflect people leaving the labor force rather than finding work.
• Over-reading one survey against the other. A single month in which household employment falls while payrolls rise is common and usually noise.
• Missing one-off distortions. Strikes, hurricanes, the timing of the 12th relative to pay periods, and government shutdowns all show up in the data. In April 2020, BLS noted that many people absent from work were recorded as employed rather than as unemployed on temporary layoff; had they been classified as BLS intended, the unadjusted unemployment rate would have been almost 5 percentage points higher.
• Average hourly earnings composition. In April 2020 average hourly earnings jumped $1.34 in a month, which BLS attributed largely to the loss of lower-paid jobs, not to pay raises. See Wage growth: average hourly earnings, real wages and composition.
• Forgetting the gap. The missing October 2025 household data break any calculation that assumes every month exists, including year-over-year changes and moving averages.

The jobs report next to other labor data

How Finabase does it

Finabase does not publish an economic calendar or a page of jobs data. The payroll and unemployment charts on this page are ours, drawn from the public BLS series on FRED as downloaded on October 30, 2026.

What the app does show is how markets price the report. The Treasury page lists ten yields from 1 month to 30 years, with the 2-year and 10-year (US10Y) among them, so you can see how a jobs number reprices expected Fed policy across maturities. The futures page carries stock-index futures such as the S&P 500 and Nasdaq contracts, which trade at 8:30 a.m. when the report comes out, before the stock market opens, plus Treasury futures; the forex page shows the dollar's moves against other currencies. For coverage of the release itself, the economy topic in Discover collects web articles, and News carries market headlines.

Frequently asked questions

What time is the jobs report released?

The Employment Situation is released at 8:30 a.m. Eastern Time, usually on a Friday early in the month after the reference month. The dates are set in advance on the BLS release schedule; for example, the October 2026 report is scheduled for Friday, November 6, 2026, and the November report for Friday, December 4. Dates can move: the November 2025 report came out on Tuesday, December 16, 2025, after the funding lapse.

What does "nonfarm" mean in nonfarm payrolls?

It means the count excludes farm workers. The establishment survey also leaves out the self-employed, private household workers and unpaid family workers, and it counts jobs rather than people. The household survey, which produces the unemployment rate, does include farm and self-employed workers.

Why are jobs numbers revised so much?

Revisions add late survey responses and, once a year, align the series with a near-complete count of jobs from unemployment-insurance tax records. The first estimate is built from the responses received by the deadline, so it is the least complete. July 2024's first print of +114,000 now stands at +53,000.

How many jobs does the economy need to add each month?

There is no fixed number. The pace that keeps unemployment steady depends on population growth and participation, which change over time, so estimates of this "breakeven" vary widely. The data show a narrower point: in 2025, payrolls grew by about 10,000 a month on average while the unemployment rate drifted up from 4.0% in January to 4.4% in December.

Why can payrolls rise while the unemployment rate also rises?

Because they come from different surveys measuring different things. Payrolls count jobs at businesses; the unemployment rate counts people. If the labor force grows faster than the number of employed people, unemployment rises even while jobs are being added, and the two surveys' sampling errors can also pull in opposite directions in any one month.

Summary

The jobs report is two surveys in one release: an establishment survey that counts jobs on payrolls, and a household survey that measures who is working, unemployed or out of the labor force. Its headline payroll change carries a 90% error band of about ±122,000, is revised twice in the following months, and is re-benchmarked once a year, so the first print is a starting estimate rather than a verdict. In the data through September 2026, payroll growth slowed from +377,000 a month in 2022 to +10,000 in 2025 and +68,000 in 2026, with frequent negative months, while unemployment stood at 4.2%. Read the report as a trend across months and across indicators, measure it against expectations, and watch how it moves the 2-year yield, because that is how it reaches every other price.