How to Read the U.S. Jobs Report: NFP and Its Traps

On the first Friday of the month, at 8:30 a.m. Eastern time, one number can move rates, the dollar and equities in a second: NFP, U.S. nonfarm payroll jobs created. It is the most traded macro data point, and one of the most poorly read. People treat it as a hard fact, when it comes from an incomplete survey, carries a margin of error almost nobody cites, and is revised the following month. Worse, two surveys coexist inside the same report and often contradict each other. This guide shows how to move from the Friday headline to a reading that holds up.

Two surveys, two stories

The jobs report is not one number but two surveys the BLS publishes together. The establishment survey, known as CES, surveys roughly 119,000 businesses and government agencies each month, covering about 622,000 worksites and around 26% of total payroll employment. It produces NFP, hours worked and wages. It counts jobs: one person holding two jobs is counted twice, while farmers and self-employed workers are excluded.

The household survey, known as CPS, surveys about 60,000 households and classifies every person aged 16 or older as employed, unemployed or not in the labor force. It produces the unemployment rate, participation rate and broader underemployment measures. It counts people, not jobs.

These two employment measures often diverge, and that is normal: one counts jobs, the other individuals. In May 2026, they fortunately pointed in the same direction, with 172,000 jobs created on the establishment side and 149,000 more employed people on the household side, while unemployment stayed at 4.3%. But in a divergent month, knowing which measure carries the relevant signal avoids confusion: for hiring momentum, use NFP; for unemployment, use the household survey.

Reading NFP, and its margin of error

NFP measures the net monthly change in nonfarm payroll employment. In May 2026, it came in at 172,000, far above consensus near 80,000, and was read as a robust labor market that pushed rate-cut expectations further away. The number includes a statistical adjustment, the birth-death model, which estimates jobs created or destroyed by firms that are born or die before appearing in the sample.

The central trap is ignored by almost every comment: NFP is an estimate, not a count. The BLS states that the 90% confidence interval for the monthly change is roughly plus or minus 122,000. In other words, the May print at 172,000 is, with 90% confidence, really somewhere between about 50,000 and 294,000. Any print below 122,000 is not statistically distinguishable from zero. On top of that, the first estimate is based on only about 60% of expected responses, versus 70% before the pandemic. The number delivered as fact on the first Friday is therefore a first approximation, based on a fraction of a survey that covers only a quarter of employment.

The May 2026 NFP print, +172,000, carries a 90% confidence interval of about plus or minus 122,000, or a real range from +50,000 to +294,000. Any print below 122,000 is not statistically distinguishable from zero. Source: BLS, Employment Situation technical note, May 2026.

Revisions, or why the first number often lies

The report is corrected in two ways, and this is where much of the signal sits. First, each month, the previous two months are revised as late responses come in. In the May report, March was revised up from 29,000 to 214,000, and April from 64,000 to 179,000, adding 93,000 more jobs than initially reported. Revisions can go both ways, and they often move more than the current month’s surprise.

Second, once a year, the establishment survey is benchmarked to near-exhaustive unemployment-insurance records. The gap between the March survey estimate and the actual count is used as an approximate measure of total error. That correction has been huge in recent years. The preliminary benchmark revision for March 2025 subtracted 911,000 jobs, or 0.6% of the total. Once the process was finalized in early 2026, 2025 job creation was cut from 584,000 to 181,000, a reduction of more than 400,000 jobs: the actual pace was only about 15,000 per month, versus 48,000 shown in real time. For much of 2025, the reported labor market was almost three times stronger than the reality later confirmed. Reading NFP always means remembering that the first print can be erased.

The benchmark revision cut 2025 job creation from 584,000 to 181,000, an actual pace of roughly 15,000 per month versus the 48,000 shown at the time. The preliminary March 2025 revision alone subtracted 911,000 jobs. Source: BLS, CES program and benchmark revision.

Unemployment: U3, U6 and the participation trap

The headline unemployment rate is U3: the number of unemployed people divided by the labor force. In May 2026, it was 4.3%. But there is a broader measure, U6, which adds discouraged workers and people forced into part-time work because they cannot find full-time jobs; it stood at 8.1%. The gap between the two says something about labor-market quality, not just volume.

The most counterintuitive trap concerns the participation rate, the labor force as a share of the total population, at 61.8% in May. The unemployment rate can fall for two opposite reasons: because unemployed people find work, which is good news, or because discouraged people stop looking and disappear from the statistics, which is bad news. A falling unemployment rate driven by weaker participation is not a sign of health. Always read unemployment together with participation: one without the other is easy to misinterpret.

In May 2026, the official U3 unemployment rate was 4.3%, but the broader U6 measure, which includes discouraged workers and involuntary part-time workers, reached 8.1%. With participation at 61.8%, the gap between the two measures says something about labor-market quality, not only its volume. Source: BLS household survey.

Wages and hours remain, often neglected. In May, average hourly earnings rose 0.3% on the month and 3.4% year over year. But against 4.2% inflation, that gain was negative in real terms: hourly purchasing power fell. The average workweek, stable at 34.3 hours, is a useful leading indicator because employers often adjust hours before headcount.

Reading the report in practice

Everything is public and free on the BLS website, on the first Friday of the month at 8:30 a.m. Eastern time; the June 2026 report is released on July 2. The BLS even publishes a table showing which household-survey changes are statistically significant, a healthy reflex before reacting. The reading method rests on a few principles. Never read NFP without its margin of error or the revisions to the previous two months. Cross-check the two surveys instead of isolating one. Read unemployment with participation. Above all, follow the trend over several months rather than the Friday surprise, because the trend, not the surprise, survives revisions.

Two cautions on reliability. The October 2025 household survey could not be collected because of the partial federal government shutdown, a gap still visible in the series. And the institutional context has become tense: in summer 2025, after disappointing numbers and large downward revisions, BLS leadership was reshuffled, feeding debate about the independence of public statistics. None of this invalidates the data, but all of it argues for reading it as what it is: an imperfect estimate, not a verdict.

The jobs report should not be read alone. It cross-checks with inflation: the guide on CPI and U.S. inflation shows why wages at 3.4% under inflation at 4.2% erode purchasing power. It weighs directly on the Fed, whose labor-market reading feeds the dot plot and SEP: the resilience of employment helped shift the median rate path toward a hike in June 2026. And positioning before the release can be read through the CFTC COT report. For the broader process, see the site’s methodology, and the economic calendar for release dates.

Read properly, the jobs report remains the best monthly X-ray of the U.S. labor market. Read badly, it becomes a weather vane: people overreact to a number whose margin of error makes it uncertain, whose next-month revision moves it, and whose benchmark revision can erase it. The signal is not in the Friday headline. It is in the trend, revisions and consistency between the two surveys. The rest is noise that markets trade anyway.

Main sources:

BLS, “The Employment Situation, May 2026”: NFP +172,000 in May, March and April revised to +214,000 and +179,000, average hourly earnings +0.3% month over month and +3.4% year over year, workweek at 34.3 hours, next release on July 2, 2026.
BLS, Employment Situation technical note: two surveys, CES and CPS, birth-death model, benchmark revision using unemployment-insurance records, 90% confidence interval for monthly NFP change of plus or minus 122,000.
BLS, Current Employment Statistics program: preliminary March 2025 benchmark revision of −911,000, or −0.6%, real average hourly earnings down 0.1% in May, publication schedule.
BLS, household survey, Employment Situation A tables: U3 unemployment at 4.3%, broad U6 at 8.1%, participation rate at 61.8%, long-term unemployed and discouraged workers.
Real Investment Advice, “BLS Jobs Report Is Broken?”: 2025 cut from +584,000 to +181,000 after benchmark, actual pace around 15,000 per month versus 48,000 reported, response rate falling toward 60%, sample covering 26% of employment.
CNBC, “Jobs report May 2026”: print well above consensus, U6 slightly lower at 8.1%, household employment up 149,000, context around BLS leadership change in summer 2025.