Published on: 2026-09-08
Updated on: 2026-09-08
The US nonfarm payrolls report gives markets one of the fastest readings of employment conditions, but the number released on NFP Friday is not the final historical count of jobs. It is the first estimate from the Bureau of Labor Statistics’ Current Employment Statistics survey, produced before every sampled employer has necessarily reported. The estimate is revised twice over the following two months, then later checked against a much broader employment count through the annual benchmark process.
These stages serve different purposes. The first print prioritises timeliness, monthly revisions refine the survey estimate for particular months, and the annual benchmark tests whether the estimated level of payroll employment has remained aligned with wider administrative records. A monthly revision changes the estimate for a particular month’s payroll change, while the annual benchmark primarily corrects the employment level against a near-census count. They are not the same kind of revision.
The first NFP print estimates payroll employment using the information available at the initial release deadline.
Each monthly estimate is revised twice as additional employer responses are incorporated.
The annual benchmark differs from monthly revisions because it re-anchors the employment level to broader population-based records rather than simply revising one month’s payroll gain.
One revision can be routine noise, while repeated revisions in the same direction can change the assessment of labour-market momentum.
Nonfarm payroll employment comes from the Current Employment Statistics, or CES, establishment survey. The CES program estimates employment, hours and earnings for workers on nonfarm payrolls. Each month, BLS surveys about 119,000 businesses and government agencies representing roughly 622,000 worksites, with the active sample covering about 26% of nonfarm payroll jobs.
For employment, the headline NFP figure measures the estimated net change in payroll jobs from one month to the next. The reference period is generally the pay period that includes the 12th day of the month, and results are published across detailed industries as well as for total nonfarm employment.
The headline monthly change is seasonally adjusted, while BLS also uses a birth-death model to estimate the net effect of new businesses and closures that cannot be fully observed in real time. The first print is designed to provide a timely reading of employment conditions, not a final census of every US payroll job.
After the first estimate is published, BLS continues to receive reports from establishments that missed the initial cutoff. Those responses are incorporated into later releases, making the estimate for the same month more complete.
| Stage | Timing | What changes |
|---|---|---|
| First preliminary estimate | Initial release | Earliest sample-based estimate |
| Second preliminary estimate | One month later | Additional employer reports included |
| Final sample-based estimate | Two months later | Further reports incorporated |
BLS publishes the first preliminary estimate roughly three weeks after the reference period. It is revised the following month and again two months after its original publication, after which it is normally held constant until annual benchmarking. Seasonal-adjustment recalculation can also contribute to the changes.
A month initially reported at 180,000 jobs, for example, could later be revised to 155,000 and then 148,000. Those figures are illustrative, but they show that revisions update the estimate of the same month rather than describe a new employment event.
A revision does not necessarily mean the initial estimate was produced incorrectly. It reflects a preliminary estimate being updated as more complete information becomes available. The latest headline and revisions to earlier months should also not be added together as though they cover one period, because each refers to a different reference month.
The annual benchmark changes the source of comparison. Monthly revisions remain within the CES survey framework and refine estimates for individual months as more establishment responses arrive. Benchmarking instead compares the CES employment level with a much broader population-based count derived mainly from state unemployment-insurance tax records collected through the Quarterly Census of Employment and Wages, or QCEW.
For the national series, BLS uses March as the benchmark month. It compares the CES sample-based employment estimate with the corresponding broader population employment count, and the difference between the two becomes the benchmark revision.
That distinction is important. A monthly revision changes the estimate of payroll growth for a particular month. An annual benchmark primarily corrects the level of employment at the March benchmark point. It should not be interpreted as simply adding or subtracting the benchmark revision from March’s monthly payroll change.
Once the benchmark is incorporated, BLS does more than make a one-month adjustment. The March sample estimate is replaced by the benchmark employment level. BLS then gradually distributes the benchmark difference backward through the preceding 11 months using a linear wedge, while subsequent months are recalculated from the corrected March anchor.
The adjustment therefore builds progressively through the period between annual benchmarks rather than being assigned entirely to March. The annual national benchmark ultimately affects 21 months of previously published not-seasonally-adjusted CES data. In addition, five years of seasonally adjusted data may be revised as seasonal-adjustment models and factors are refreshed.
A preliminary benchmark does not immediately replace the official payroll series. On August 28, 2026, BLS estimated that the March 2026 benchmark revision to total nonfarm employment would be −79,000 jobs, or −0.1%, while the preliminary revision to private employment was −178,000. BLS said the absolute average annual benchmark revision over the previous 10 years was 0.2% of total nonfarm employment.
A −79,000 benchmark revision does not mean BLS discovered that 79,000 fewer jobs were created during March, nor does it mean 79,000 will simply be subtracted from March’s monthly payroll gain. It means the broader benchmark count indicates that the level of total nonfarm payroll employment at the March 2026 benchmark point was 79,000 lower than the CES sample-based estimate had indicated.
When the benchmark is incorporated, that March employment level becomes the new anchor for the series. The discrepancy is progressively wedged backward across the period following the previous March benchmark, reaching the full adjustment at the new March benchmark point, while subsequent estimates are recalculated from that corrected level.
The March 2026 preliminary benchmark will be incorporated only when the final benchmark is published with the January 2027 Employment Situation report in February 2027.
Routine revisions are expected because more information becomes available after the first release. Larger adjustments are more likely when economic conditions are changing quickly or when the early survey is harder to reconcile with the broader employer population.
Economic turning points can make employment changes harder to estimate because hiring, layoffs, business formation and closures may shift faster than recent historical relationships suggest. New firms also do not enter the survey frame immediately, while failed businesses can be identified with a lag.
Sampling variation, unusual response patterns and seasonal adjustment can add further uncertainty. For annual benchmarking, the gap can widen when the CES survey gradually drifts above or below the broader administrative employment count.
A large benchmark revision should therefore not automatically be attributed to one model or one statistical issue. It reflects the difference between two independently derived employment measures, each with its own limitations.
The importance of a revision depends on both its size and its pattern. One downward adjustment can be routine statistical noise, while repeated downward revisions across several months can suggest that hiring momentum was weaker than initially reported. Persistent upward revisions can point in the opposite direction.
Sector composition also matters. A revision concentrated in one industry carries a different signal from broad changes across several cyclical sectors. Benchmark revisions answer another question altogether: whether the estimated level of payroll employment was above or below the broader administrative count.
| Question | Most relevant data |
|---|---|
| What new employment signal arrived today? | First NFP print |
| Was an earlier month stronger or weaker? | Monthly revision |
| Is recent hiring momentum changing? | Several prints plus revisions |
| Was the employment level misestimated? | Annual benchmark |
For the immediate market reaction, the first print usually carries the greatest weight because it is new information and is compared directly with consensus expectations. Treasury yields, the US dollar and equity-index futures can respond quickly when payrolls, unemployment or wages surprise.
For a broader assessment of labour-market momentum, the latest print is better read alongside revisions to the previous two months, the three-month average, industry breadth, unemployment, average hourly earnings and hours worked. A strong headline accompanied by repeated downward revisions can present a weaker picture than the latest number alone suggests.
The annual benchmark operates on a longer horizon. It is usually less important for the immediate market reaction, but more useful when assessing whether the historical payroll level and employment trend were measured accurately. Because it re-anchors the employment level rather than simply changing one month’s payroll gain, it should be interpreted differently from routine monthly revisions.
The first NFP print, monthly revisions and annual benchmark serve different purposes. The first print provides the earliest survey-based signal, the next two releases refine individual monthly estimates as more employer reports arrive, and the annual benchmark checks the employment level against a much broader administrative count.
For market interpretation, the first print matters most for the immediate surprise, while the direction of subsequent revisions helps show whether that initial signal holds up. The annual benchmark provides the broader historical check by re-anchoring the employment level and propagating that correction through the historical series, making revisions part of the labour-market story rather than an afterthought.
